O PEN A CCESS TO T ELECOMMUNICATIONS
I NFRASTRUCTURE AND D IGITAL S ERVICES :
C OMPETITION , C OOPERATION AND R EGULATION
Zur Erlangung des akademischen Grades eines
Doktors der Wirtschaftswissenschaften
(Dr. rer. pol.)
von der Fakultät für
Wirtschaftswissenschaften
am Karlsruher Institut für Technologie (KIT)
genehmigte
DISSERTATION
von
Daniel Schnurr, M. Sc.
Tag der mündlichen Prüfung: 25. Juli 2016
Referent: Prof. Dr. Jan Krämer
Korreferent: Prof. Dr. Karl-Martin Ehrhart
Karlsruhe, 2016
Abstract
Open Access, defined as the non-discriminatory access to an upstream bottleneck resource, takes a central role in information and communications technology markets
with its direct and indirect ramifications on competition and innovation. Moreover,
recent technological innovations have fundamentally affected i) the market structure
and regulatory paradigm at the infrastructure layer, where Open Access safeguards access to communications access networks, and ii) the evolution of digital services markets, where Open Access is discussed with regard to immaterial upstream bottleneck
resources such as data or intellectual property.
This thesis investigates the competitive and cooperative interactions in these markets,
where firms require access to an essential input resource. Thereby, theoretical analyses
and experimental evaluations are employed to examine market outcomes under alternative regulatory institutions. In particular, experiments in continuous time are utilized
to allow for oligopoly competition with asynchronous strategic interaction. Serving as
a testbed for policy proposals, the thesis aims at identifying welfare implications of
designated regulatory institutions and informing the design of new institutions prior
to implementation in the field.
Based on a unified framework for Open Access regulation, margin squeeze regulation,
which has recently been emphasized as an alternative to wholesale price regulation
in the European Union regulatory framework for telecommunications infrastructure,
is scrutinized. The theoretical and empirical analyses in this thesis demonstrate that
in the case of infrastructure competition, margin squeeze regulation may benefit nonintegrated retailers in specific cases, but never benefits consumers. Thus, in contradiction to the rationale underlying the current legislative implementation in Europe, com-
i
peting infrastructures do not represent exogenous competitive constraints, but increase
retail prices strategically in anticipation of the regulatory rules.
In the case of wholesale competition, access for a retailer may be provided competitively by two vertically integrated firms. Remarkably, in a continuous time economic
laboratory experiment, wholesale and retail prices are found to be higher under wholesale competition than under a wholesale monopoly. The finding that consumers are
worse off under competition is rationalized by an investigation of collusion incentives.
Moreover, a complementary price commitment rule is found to substantially reduce
tacit collusion. A validation study with expert participants provides support that experimental outcomes are indeed similar to the student sample. These results point to
a dilemma of the Open Access rationale: whereas non-discrimination may provide the
basis for competition on equal terms, symmetry may facilitate coordinated behavior
among competitors to the detriment of consumers.
The effect of the number of firms on tacit collusion, a central concern in merger control
proceedings, is further examined by means of an empirical meta-analysis and two experimental studies. It is shown that tacit collusion decreases strictly with the number
of competitors in industries with two, three and four firms. Although previous experimental studies could not affirm that tacit collusion is higher in markets with three than
with four firms, evidence for this fact can be provided for symmetric and asymmetric
firms, and under Bertrand and Cournot competition.
Finally, voluntary access relationships in digital services markets are explored. To this
end, the competitive effects of social logins that allow for the sharing of user and usage
data between online content providers and a social network are analyzed theoretically.
It is shown that content providers may adopt a social login even if this strategic decision
makes them ultimately worse off, i.e., they find themselves in a prisoner’s dilemmalike situation. Thus, market failures may occur due to the discriminatory access by the
social network. In this vein, voluntary access may be offered by a dominant gatekeeper
to protect its position, albeit not through exclusion, but through exploitation. This calls
for consideration of Open Access institutions at the digital services layer.
ii
Acknowledgements
Foremost, I would like to thank Prof. Dr. Jan Krämer who has not only been an outstanding mentor, but also has become a dear friend. I am particularly grateful for his
constant motivation and inspiration as well as the numerous insightful discussions.
I am also indebted to Prof. Dr. Christof Weinhardt who offered me the opportunity and
the freedom to pursue my research interests at the Institute of Information Systems and
Marketing (IISM). Moreover, I would like to thank my co-advisor Prof. Dr. Karl-Martin
Ehrhart and Prof. Dr. Ingrid Ott, who have also served on my thesis committee, for
their time spent on reading this thesis and insightful comments on my work.
My sincere appreciation goes to all my colleagues at the IISM, FZI and the University
of Passau. Not only have the discussions been a source of inspiration to me, but the
enthusiasm and established friendships have created a joyous place to work at. In particular, I thank all members of the TELCO research group, Niklas Horstmann, Philip
Köhler, Lukas Wiewiorra and Michael Wohlfarth for close research collaborations and
a unique team spirit. It has been a great run.
Finally, I would like to thank those people without whom this work would not have
been possible. I thank my parents Hildegard and Wilhelm and my sister Anja for their
unconditional love and support throughout my life, and Jacqueline for her love, encouragement and the joy she brings to my life. I owe them more than words can describe.
iii
Preface
I
N the European telecommunications sector, the Open Access (OA) concept represents a well-known cornerstone of the regulatory framework, established during the
liberalization and privatization of the industry (Farrell and Weiser, 2003). Guided by
the Ladder of Investment rationale (Cave and Vogelsang, 2003; Cave, 2006a), access and
price regulation has been implemented to foster competition and investments by entrants such that regulation could be lifted stepwise and ultimately as a whole. Fixed
and mobile telecommunication markets quickly grew in the 1990s and early 2000s,
while entrants relying on regulated access to the incumbent’s network were able to capture significant market shares. However, these trends have slowed down significantly
or have even been reversed more recently (European Commission, 2014b). Moreover,
technological advancements in this period have tremendously improved the technical
capabilities of the network infrastructure, but have also fundamentally altered the provisioning of services on top of these networks. In particular, digital convergence, i.e., the
implementation of the Internet Protocol (IP) as a uniform network standard in combination with a layered protocol architecture, modular composition of end-to-end connections, and infrastructure-transparent services, has laid the foundation for distributed
digital services and a globally connected information system.
In consequence, the telecommunications infrastructure today serves a much more general class of digital information services rather than only communications services. On
the one hand, the success of these online information services has direct implications
for telecommunications providers’ business models and possible cooperative strategies. In particular, as complementary goods they increase demand for access services
at the network layer. Yet, in contrast, as substitutes they represent additional competitive constraints, especially in the case of communications services. On the other
hand, access networks may be affected indirectly by the increased public attention and
ensuing policy considerations due to their role as gateways to the Internet and to the
respective digital services (see, e.g., the public debate on net neutrality summarized by
Krämer et al., 2013). More specifically, the following stylized trends summarize how
business models and competitive strategies in the telecommunications markets have
been altered by technological progress and adjusted policy considerations. As will be
discussed subsequently, these trends question traditional regulatory practice and the
outlines of the original sector-specific regulatory framework with the advent of nextgeneration access networks (NGANs). Although not complete, the listed trends call for a
v
Preface
reconsideration of the role, the objectives and the design of regulatory OA regimes at
the telecommunications infrastructure layer.
Trends at the telecommunications infrastructure layer
1) Infrastructure competition: Based on the IP standard, different next-generation access
(NGA) platforms, which have initially been built for distinct services such as telephony
and cable television, are able to deliver digital communications and information services to consumers at fixed locations (De Bijl and Peitz, 2008). In consequence, ensuing infrastructure competition challenges the traditional notion of the fixed local access
network as a natural monopoly (Vogelsang, 2013). In addition, different geographical
footprints of those networks in combination with the emergence of regional operators
have led to a situation where the number of competitors differs geographically (Bourreau et al., 2015). Therefore, in densely populated areas, consumers can choose from
several operators, which are each in possession of their own distinct network infrastructure. The market structure in fixed telecommunications thus starts to resemble the
market structure in mobile telecommunications, where a low number of infrastructure
competitors compete strategically at the retail level and possibly at the wholesale level.
Thus, wholesale competition on the basis of infrastructure competition may facilitate
voluntary wholesale access for independent retailers and virtual network operators
(Ordover and Shaffer, 2007; Bourreau et al., 2011). The entry of network operators relying on a distinct infrastructure may alleviate concerns about exploitation of market
power by a single dominant firm, but instead, tacit collusion, i.e., the implicit coordination of firms’ behavior to the detriment of consumers, may become the central concern
of regulators and antitrust authorities in an oligopolistic market structure (Parker and
Röller, 1997; Ivaldi et al., 2003).
2) Next-generation access technologies: The availability of new high-bandwidth technologies, in particular fiber optics (Kazovsky et al., 2007; Wong, 2012), as well as the development of upgrade solutions for existing technologies, such as Vectoring (Broadband
Forum, 2012) and Fiber to the Distribution Point (Broadband Forum, 2015), allow for
transmission speeds of several orders of magnitude higher than legacy communications network technologies. At the same time, growing demand for new digital services
and continued digitization of traditional physical goods industries reinforce the need
for the rollout of NGANs based on these technologies (European Commission, 2016b).
With regard to the policy objectives that guide regulatory intervention at the European
Union (EU) level, these requirements have begun to shift the focus from competition to
a dual emphasis on both competition and investments (see, e.g., Kroes, 2012).
vi
3) Virtual network operators: The existence of different NGA technologies in practice
increases the heterogeneity of potential wholesale access products across geographic
locations (Bundesnetzagentur, 2011). At the same time, virtualization at the network
layer allows for the specification of harmonized access products across technologically
heterogeneous infrastructures, which may serve as the technological foundation for
(cross-border) competition in a European digital single market (European Commission,
2013e, 2015a). In combination, both trends have enabled new operator models and
access offers at different levels of the fixed infrastructure and network operations value
chain (Dewenter and Haucap, 2006; Banerjee and Dippon, 2009). Thus, multiple virtual
network operators and therefore competition may be sustained on top of few or even a
single infrastructure.
4) Fixed-mobile integration: Digital convergence at the network layer encompasses not
only fixed networks, but also promotes fixed-mobile integration, i.e., the technological integration of fixed and mobile telecommunications infrastructures as well as the
bundling of respective communications services (OECD, 2015). From a supply-side
perspective, increasing wireless transmission speeds and ensuing higher data volumes sent over radio interfaces require the integration of base stations into dense
and high-bandwidth concentration networks (5G PPP, 2015). At the same time, consumers’ demand for communications services bundles including a seamless handover
between fixed and mobile services has significantly grown (Grzybowski and Liang,
2014). In consequence, the traditional technological boundaries between fixed and
mobile telecommunications markets blur, which is further indicated by hybrid access
products that connect to a fixed as well as a mobile access network, simultaneously
(Leymann et al., 2015). In conclusion, these developments challenge the notion of delineated regulatory regimes dedicated to specific infrastructures, such as mobile and fixed
networks, and ultimately suggest the necessity of regulatory convergence irrespective of
the underlying technology. Moreover, the consolidation in the mobile telecommunications industry has effectively reduced the number of distinct infrastructures (Genakos
et al., 2015). In consequence, the market structure is becoming increasingly similar to
fixed telecommunications markets as noted above.
5) Public-sector participation: The increasing economic and societal relevance of digital
services has led to direct public intervention at the network infrastructure and operations level, especially in rural less densely populated areas. In particular, consideration
of spill-over effects and positive externalities, the universal service principle, and the
objective to end the digital divide are cited as legitimatory reasons for those state activities (see, e.g., Gómez-Barroso and Feijóo, 2010, for an overview). If public-sector participation complements private activity, most notably in the case of state aid, the effects and
trade-offs resulting from mandatory access obligations are magnified: Whereas public
intervention strengthens the rationale for non-discriminatory access offers to wholesale services in order to avoid the distortion of competition, such access obligations
vii
Preface
may discourage private investments particularly in those areas where they are most
required.
6) Over-the-top communications services: The technical decoupling and logical separation
of network operations has enabled IP-based over-the-top services providers to enter
communications and digital services markets without an own infrastructure. In turn,
this diminishes returns for network operators from traditional revenue sources (Peitz
and Valletti, 2015). Yet, at the same time, these services are the main drivers of demand
for fixed and mobile access services, which however may be difficult to monetize as
illustrated by objections raised in the context of net neutrality (Krämer et al., 2013).
In consequence, authorities are challenged to decide whether the regulatory regime at
the network access layer needs to be adjusted in light of this additional competitive
pressure at the services layer in order to create a level playing field (cf. Krämer and
Wohlfarth, 2015).
In consequence, these trends raise several issues with regard to the design and implementation of regulatory institutions that can safeguard OA at the network infrastructure level. In fact, these issues are currently controversially debated with regard to
a realignment of the European regulatory framework (European Commission, 2016c).
Whereas price regulation in retail markets has been faded out from the European Commission’s (2014a) recommendation on relevant markets susceptible to ex ante regulation, price regulation of wholesale prices for physical and virtual access products remains the recommended default and the prevalent regulatory remedy in European
markets up to date. However, with respect to the rollout of NGA technologies, some national regulatory authorities (NRAs) have attempted to depart from the commission’s
recommendation and suggested alternative regulatory regimes in lieu of cost-based access regulation. For instance, the Austrian NRA proposed wholesale access prices on
the basis of a margin squeeze test (European Commission, 2013f), whereas the Spanish NRA exempted NGANs above a defined data transmission threshold from price
regulation (European Commission, 2008). This latter approach may be viewed as a
form of regulatory holidays, which has been proposed to resolve the truncation problem in the case of risky investment (Gans and King, 2003, 2004). Despite the concerns
about insufficient investment incentives due to cost-based price regulation, the European Commission in the past has critically scrutinized and subsequently often appealed
those proposals on the ground that they would contradict the harmonized regulatory
framework at the EU level (see, e.g., European Commission, 2013d). Instead, European policy makers frequently argue that static and dynamic incentives are invariably
aligned and reinforcing each other (see, e.g., Vestager, 2015), despite growing empirical
evidence (inter alia Briglauer et al., 2013; Bacache et al., 2014; Klumpp and Su, 2015)
that suggests the presence of a trade-off between static and dynamic efficiency.
viii
Recently, several high-profile policy initiatives and legislative proposals point to an increasing willingness at the European level to rethink fundamental cornerstones of the
regulatory framework as the aforementioned trends become more noticeable and urgent (Bauer, 2010; Ünver, 2015). The Europe 2020 strategy, adopted in 2010, with its Digital Agenda for Europe has emphasized the need to roll out high-speed NGANs and set
ambitious coverage and adoption targets (European Commission, 2010, 2014b). Thereupon, the European Commission’s (2013a) recommendation on non-discrimination and
costing methodologies marked the first executive action at the European level diverging from the prevalent paradigm of ex ante wholesale price regulation, as the commission explicitly enabled national regulators to replace cost-based price regulation with
non-discrimination obligations. In the same year, the European Commission (2013e)
presented its Connected Continent legislative package, which aimed at strengthening the
European digital single market, offering more favorable conditions on market consolidation and suggested, among others, the transition to simplified and uniform (virtual)
wholesale products across EU member states. However, the final political compromise
reached in 2015 only included new rules on international mobile roaming charges and
net neutrality (European Commission, 2015d), whereas the revision of regulatory rules
on wholesale access was postponed (see Renda, 2015, for a commentary). The latest
initiative by the European Commission (2015a), the Digital Single Market strategy reiterates the need for a framework aimed at strengthening investment incentives, but has so
far not discussed any legislative changes in this regard (European Commission, 2015c).
Yet, at the latest, the upcoming review of the European Regulatory Framework will require a reevaluation and possibly a redesign of regulatory rules on wholesale access at
the European and the member states level (European Commission, 2016c).
In this context, sector-specific regulation may not be viewed and evaluated in isolation,
but should be considered within the larger competition policy and antitrust framework.
Most notably, recent decisions in merger control proceedings (Genakos et al., 2015) illustrate the interdependency and interaction between both frameworks with regard to
the effects on competitiveness of oligopolistic markets. In particular, the decision on
the allowed level of consolidation and thus the number of competitors in a market has
direct ramifications on the necessity and adequacy of regulatory remedies (see, e.g.,
the discussion of merger consequences for the definition of regulatory markets by Bundesnetzagentur, 2015). Moreover, antitrust rules may constrain the scope of cooperative
agreements among competitors, which may be conducive to investment incentives, but
may also be to the detriment of competitors or consumers (Jorde and Teece, 1990; Baumol, 2001; Schellingerhout, 2011). Finally, ex ante regulatory mechanisms and ex post
competition policy rules need to be aligned and applied consistently in particular with
respect to the enforcement of anti-competitive conduct prosecuted under both frameworks (see, inter alia, Geradin and O’Donoghue, 2005; Hellwig, 2008).
ix
Preface
Trends at the digital services layer
Above all, recent policy reform initiatives in the information and communications technology (ICT) context are influenced to a large extent by the developments at higher layers of the value chain: the emergence of new digital services markets and the success
of online platform business models, which in turn have allowed some of these platform operators to gain significant market power in their respective markets (Peitz and
Valletti, 2015). Thus, technological progress did not only transform the infrastructure
layer, but enabled the development of new services markets on top of it. In particular,
the virtualization of goods through digitization together with the logical separation of
the services layer from the infrastructure layer have lowered supply-side barriers to
entry into services markets. Moreover, they have allowed for rapid scalability of firms’
operations due to zero marginal costs, which in turn facilitate bundling of services and
innovative pricing strategies (see, inter alia, Bakos and Brynjolfsson, 1999; Shampanier
et al., 2007). In addition, low distribution costs allow to serve content and services at a
global level and exploit substantial economies of scale. Although these factors, among
others, have induced the market entry of a vast number of specialized online content
and services providers and thus have significantly increased the accessible variety of
goods and services for consumers globally (see, e.g., Brynjolfsson et al., 2003), network
and feedback effects together with scale advantages have facilitated the emergence of
a few dominant firms (see, inter alia, Evans and Schmalensee, 2009, 2013). In consequence, these firms have been characterized as new gatekeepers in the ICT value chain
(Baye and Morgan, 2001; Ballon and Van Heesvelde, 2011). Moreover, among these
firms, which have first emerged in a particular market, there is a continuing trend to
expand into adjacent markets, thus creating services ecosystems that encompass several
layers of the ICT value chain (Monopolkommission, 2015). In the light of the abovementioned developments, several of those firms have been confronted with antitrust
or regulatory scrutiny. Most notably, Alphabet (Google) with its online search and its
mobile operating system Android (European Commission, 2015b, 2016a; Nicas, 2016),
Amazon with its online retail store and its ebook platform (Budzinski and Köhler, 2015),
as well as Facebook with its social network market (Bundeskartellamt, 2016) have been
accused of abusing or leveraging their market power. At the core of these accusations,
data, more specifically user and usage data, is often referred to as the new bottleneck
resource in ICT markets (Graef et al., 2015).
Data may create economic value as the enabler of new and enhanced marketing instruments as well as a vital input factor for the creation and refinement of information
services. On the one hand, data-driven business models allow for indirect monetization strategies either via two-sided pricing models, in particular advertising (see, e.g.,
Evans, 2008), or via price discrimination (see, e.g., Taylor, 2004; Acquisti and Varian,
2005). The exploitation of indirect network effects in the former case allows firms to
x
offer services to consumers possibly at a zero price. Moreover, the ability to employ
penetration pricing and thereby to subsidize a particular market side can accelerate the
diffusion of new services and thus facilitate entry into new markets (Jiang and Sarkar,
2009). In ICT markets, where services regularly represent information goods and where
quick innovation cycles are prevalent, the former ability may prove to be critical for a
firms’ success. On the other hand, (personal) data may represent an essential supplyside input to provide and improve digital online services (cf. Levin, 2013). For instance,
online search algorithms exploit the collected data on user queries and usage traffic in
order to improve the ranking of organic search results as well as the display of sponsored search advertising (cf. Newman, 2014). Moreover, personalized customization of
services, a key differentiation parameter in digital services competition, depend directly
on the available data basis and data quality (cf. Acquisti and Varian, 2005; Thirumalai
and Sinha, 2013; Aguirre et al., 2015). Therefore, the access to data is viewed as a crucial
factor, not only for firms to succeed in ICT markets, but also for a policy framework that
reaches beyond the infrastructure level to include online markets (European Commission, 2015c). In consequence, this has reinvigorated the debate about OA, now in the
context of an upstream resource deemed essential for digital (online) ICT services (see,
e.g., Argenton and Prüfer, 2012). In particular, questions arise with regard to whether
OA concepts should and can be transferred to the services layer, as some have already
called for the regulation of OA to virtual facilities (see Krämer and Wohlfarth, 2015, for
a discussion).
However, as shown above, implications drawn from the infrastructure layer may not be
directly applicable to the services layer, as they differ with respect to several important
economic characteristics. In particular, externalities in the context of multi-sided platforms and indirect monetization models may significantly alter competitive strategies
(cf. Haucap and Heimeshoff, 2014). Thus, traditional approaches to determine market
power and anti-competitive conduct, established in the context of one-sided markets,
may be inadequate to assess the competitiveness of these markets and to finally judge
consumer harm (Wright et al., 2004). Moreover, new incentives for cooperation may
arise together with innovative sharing mechanisms with regard to data as a digital input good, that is naturally different from physical input goods, which are characterized
by large sunk cost and rivalry in use. Therefore, it is a priori unclear, whether marketdriven incentives may be sufficient to safeguard access to essential input goods as voluntary sharing agreements may even be agreed upon between competitors (see, e.g.,
Mantena and Saha, 2012). On the contrary, voluntary cooperative relationships themselves may have detrimental welfare effects for competitors and consumers (Verdier,
2013), or even for the cooperating firms themselves (see, e.g., Mantovani and RuizAliseda, 2015). Furthermore, voluntary access offers on the provider’s discretion may
be deemed insufficient in specific cases if non-discrimination is viewed as an essential criterion. Thus, non-discrimination or neutrality obligations, which have been discussed extensively at the interface between ICT infrastructure and services, may also
xi
Preface
be considered at the services layer (cf. Easley et al., 2015). In conclusion, these considerations indicate that at first a better understanding of the incentives and implications of
access relationships at the services layer is necessary in order to evaluate the need and
role for OA at this layer of the ICT value chain.
xii
Contents
Preface
v
1
Introduction
1.1 Research questions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.2 Structure of this thesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1
3
8
2
A Unified Framework for Open Access Regulation
2.1 The concept of Open Access . . . . . . . . . . . . . . . . . . . . .
2.1.1 Open Access in the context of public-sector participation
2.1.2 Voluntary Open Access . . . . . . . . . . . . . . . . . . .
2.1.3 The Open Access network model . . . . . . . . . . . . . .
2.1.4 A conceptual Open Access framework . . . . . . . . . . .
2.2 Literature survey . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.2.1 Vertical integration and separation of the access provider
2.2.2 Co-investment . . . . . . . . . . . . . . . . . . . . . . . .
2.2.3 Public-sector participation . . . . . . . . . . . . . . . . . .
2.3 Policy guideline . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
11
14
15
16
16
18
22
23
34
38
45
53
3
Theoretical Foundations and Experimental Methodology
3.1 Microeconomically founded research in Information Systems .
3.1.1 Theory as a set of models . . . . . . . . . . . . . . . . .
3.1.2 Models as mediators between theory and reality . . . .
3.1.3 Empirical analyses as the means to evaluate theory . .
3.1.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . .
3.2 Theoretical modeling and experimental evaluation . . . . . . .
3.3 Oligopoly competition in continuous time . . . . . . . . . . . .
3.3.1 Timing in experiments . . . . . . . . . . . . . . . . . . .
3.3.2 Experimental framework . . . . . . . . . . . . . . . . .
3.3.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.3.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
57
58
60
62
64
66
67
77
79
86
94
99
4
Infrastructure Competition and Open Access Regulation
4.1 Theoretical framework . . . . . . . . . . . . . . . . . .
4.2 Competitive retail fringe . . . . . . . . . . . . . . . . .
4.3 Simultaneous retail pricing . . . . . . . . . . . . . . .
4.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
103
106
107
111
114
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
xiii
Contents
5
Wholesale Competition and Open Access Regulation
5.1 Related literature . . . . . . . . . . . . . . . . . . . .
5.1.1 Wholesale monopoly and competition . . . .
5.1.2 Margin squeeze regulation . . . . . . . . . .
5.2 Experimental framework . . . . . . . . . . . . . . . .
5.2.1 Theoretical model . . . . . . . . . . . . . . . .
5.2.2 Design . . . . . . . . . . . . . . . . . . . . . .
5.2.3 Procedures . . . . . . . . . . . . . . . . . . . .
5.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . .
5.3.1 Main study . . . . . . . . . . . . . . . . . . .
5.3.2 Validation study . . . . . . . . . . . . . . . .
5.4 Repeated game analysis of tacit collusion incentives
5.5 Discussion . . . . . . . . . . . . . . . . . . . . . . . .
6
Number Effects and Tacit Collusion in Oligopolies
6.1 Review and meta-analysis of the experimental literature . . . . .
6.1.1 Experimental designs . . . . . . . . . . . . . . . . . . . . .
6.1.2 Measuring competitiveness as the degree of tacit collusion
6.1.3 Results of the meta-analysis . . . . . . . . . . . . . . . . . .
6.1.4 Discussion of the meta-analysis . . . . . . . . . . . . . . . .
6.2 Experiment with symmetric firms . . . . . . . . . . . . . . . . . . .
6.2.1 Experimental design . . . . . . . . . . . . . . . . . . . . . .
6.2.2 Procedures . . . . . . . . . . . . . . . . . . . . . . . . . . . .
6.2.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
6.3 Experiment with asymmetric firms . . . . . . . . . . . . . . . . . .
6.3.1 Experimental design and procedures . . . . . . . . . . . .
6.3.2 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
6.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
7
Voluntary Open Access in Digital Services Markets
7.1 Related literature . . . . . . . . . . . . . . . . . . . . . . . . .
7.2 A model of coopetition in online advertising markets . . . .
7.2.1 Competitive setting . . . . . . . . . . . . . . . . . . . .
7.2.2 Details of the model . . . . . . . . . . . . . . . . . . .
7.3 Competitive effects of the social login and market outcomes
7.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
8
Conclusion
213
8.1 Summary and implications . . . . . . . . . . . . . . . . . . . . . . . . . . 214
8.2 Limitations and outlook . . . . . . . . . . . . . . . . . . . . . . . . . . . . 219
A Theoretical Analyses
A.1 Oligopoly competition . . . . . . . . . . .
A.2 Margin squeeze regulation . . . . . . . . .
A.3 Wholesale competition . . . . . . . . . . .
A.4 Coopetition in online advertising markets
xiv
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
119
123
123
126
128
128
133
135
137
137
145
147
150
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
153
156
157
158
160
165
167
167
170
170
174
174
176
180
.
.
.
.
.
.
.
.
.
.
.
.
185
188
193
193
195
200
209
.
.
.
.
.
.
.
.
.
.
.
.
227
227
236
243
249
Contents
B Statistical Analyses
B.1 Oligopoly competition in continuous time . . . . . . . . . . . . . . . .
B.2 Wholesale competition and Open Access regulation . . . . . . . . . .
B.3 Number effects and tacit collusion in oligopolies . . . . . . . . . . . .
259
259
260
266
C Experimental Instructions
C.1 Oligopoly competition in continuous time . . . . . . . . . . . . . . . .
C.2 Wholesale competition and Open Access regulation . . . . . . . . . .
C.3 Number effects and tacit collusion in oligopolies . . . . . . . . . . . .
273
274
278
284
References
289
List of Abbreviations
343
List of Figures
345
List of Tables
347
xv
Chapter 1
Introduction
C
ONTINUOUS innovation and ubiquitous diffusion has put information and
communications technology (ICT) at the center of society and the economy of
the 21st century (Brynjolfsson and Hitt, 2000; Czernich et al., 2011; Cardona et al., 2013).
Numerous revolutions have been proclaimed to describe the transformations that the
ICT sector itself has undergone in response to technological progress, new business
models, and adjusted policy objectives (see Steinbock, 2005; Noam, 2010; Vogelstein,
2013; Kitchin, 2014; Jorgenson and Vu, 2016, for only a few examples). Induced by technological innovations, the outcomes of these processes are fundamentally shaped by
firms’ competitive and cooperative relationships and thus by the regulatory framework
that governs these interactions. At the same time, competition, cooperation, and regulation themselves constitute the key determinants of future innovation (Dosi, 1988). In
this context, Open Access (OA) to communications infrastructure and digital information goods, i.e., the non-discriminatory access to an upstream bottleneck resource, takes
a central role with its direct and indirect ramifications on competition and innovation
(OECD, 2013). In this regard, OA regulation may represent a major design variable to
shape market institutions and to influence and balance trade-offs between static and
dynamic effects in ICT markets (Klumpp and Su, 2010, 2015). While digitization has,
on the one hand, significantly lowered the costs to provide access, it has, on the other
hand, increased the scope and variety of potential access mechanisms. In consequence,
1
Chapter 1 Introduction
existing OA institutions may need to be rethought and redesigned. Moreover, novel
OA issues may arise in the context of newly emerging bottleneck resources and even
in cases where voluntary access relationships already exist. In particular, recent technological and policy trends have fundamentally affected i) the market structure and
regulatory paradigm at the infrastructure layer, where OA safeguards access to communications access networks, and ii) the evolution of digital services markets, where
OA is discussed with regard to the non-discriminatory access to immaterial upstream
bottleneck resources such as data or intellectual property.
The ongoing policy debates caused by the technological advancements in ICT markets
highlight the interplay between technical and legal determinants on the one hand, and
strategic processes and economic outcomes on the other hand. In order to design institutions, which can govern and shape competitive and cooperative processes according
to given policy objectives (North, 1991; Roth, 2002), it is paramount to understand, explain and predict the incentives and trade-offs that arise in these markets (cf. Gregor,
2006). Thereby, the virtual nature of digital goods allows for a new degree of freedom
with respect to the dimension, the type and the degree of possible policy interventions.
In consequence, this emphasizes the need for theory-informed market design and economic engineering (cf. Gimpel et al., 2008). As goods and services themselves become
artificial and subject to deliberate design decisions, so do the institutions that encompass them. Given the vital importance of technological and economic factors in the ICT
context as well as the interactions among them, research at the interface between Information Systems (IS) and Industrial Organization (IO) is particularly suited to establish
and refine robust theories that can guide the conceptualization and implementation
of these design proposals together with their respective policy implications. In general,
this i) requires an analysis of the incentives, externalities and trade-offs that arise from a
particular design in a particular context, and ii) calls for empirical testing of theoretical
predictions and prototypical evaluation of design proposals prior to implementation
at large scale in practice. Whereas the first issue can be addressed by analytic models
based on stylized facts and the extant theoretical body of knowledge, the second issue
requires an empirical testbed that can capture behavioral effects and isolate causal re-
2
1.1 Research questions
lationships by controlled variation, which can be achieved by laboratory experiments.
This thesis will draw on both methodologies to address the research questions outlined
in the following.
1.1 Research questions
Across different ICT markets, this thesis examines firms’ strategic incentives to provide
access to an internal upstream resource and investigates the implications for competition and consumers. Therefore, this thesis focuses on conditions that facilitate voluntary access agreements, but also identifies regulatory institutions that enforce the principles of non-discriminatory access according to the core idea of OA if market mechanisms are found to fail in this respect.
At first, however, it is necessary to clarify and define what is commonly referred to as
OA. Although the term has been widely used in the academic literature (e.g., Farrell and
Weiser, 2003; Forzati et al., 2010) and in practice (e.g., European Commission, 2009b),
varying concepts of OA have explicitly or implicitly been used and different aspects
have thereby been emphasized. Therefore, a review of suggested definitions, application scenarios and the scientific literature is crucial to establish a common understanding and provide a foundation for successful implementation in practice. Moreover, a
structured policy framework classifying relevant determinants of OA applications and
identifying alternative regulatory OA regimes is necessary to inform the current debate
about OA to next-generation access networks (NGANs) with regard to the available
policy options. As policy makers and regulatory agencies are challenged to decide between alternative regulatory regimes and to design a new regulatory framework in the
context of NGANs, the relevant economic trade-offs and the implications for different
efficiency measures identified in the economic literature need to be made transparent
in order to secure the rollout and allocation of the telecommunications infrastructure,
which is deemed vital for modern economies and societies (Czernich et al., 2011). In
summary, these considerations lead to the first research question:
3
Chapter 1 Introduction
R ESEARCH Q UESTION 1. What defines Open Access given the definitions offered by various
stakeholders in the context of telecommunications infrastructure markets? To which extent can
different regulatory regimes achieve Open Access and which trade-offs arise for the decision
between those regimes?
Margin squeeze regulation (MSR) has been proposed as an alternative regulatory institution to traditional wholesale price regulation to achieve OA to an upstream resource
supplied by a vertically integrated firm for independent retailers (European Commission, 2013a). Under such a regulatory regime, the wholesale price of a vertically integrated access provider may not exceed its retail price. Whereas theoretical analyses of
MSR exist with regard to a single vertically integrated bottleneck supplier (e.g. Jullien
et al., 2014), there is no examination of the welfare implications in the context of infrastructure competition, which is now found in many European mobile and fixed telecommunications markets, with the single notable exception of Höffler and Schmidt (2008).
Therefore, it is unclear whether MSR can safeguard competitors and consumers against
foreclosure of independent retailers that rely on the access to the upstream good. Given
its relevance as an ex ante regulatory remedy as well as a stand-alone antitrust abuse in
ex post competition law, a further empirical evaluation of theoretical predictions seems
warranted in order to attain robust findings over and beyond analytic models. Next to
its application in the case of infrastructure competition with a single access provider,
welfare implications of MSR may change under wholesale competition, which is the
case in most European mobile telecommunications markets. Yet, no study has investigated MSR in the context of wholesale competition. Thus, the second research question
asks:
R ESEARCH Q UESTION 2. What are the welfare implications of margin squeeze regulation
in the presence of several independent infrastructures that may allow for competition at the
wholesale level in addition to competition at the retail level?
More generally, the emergence of infrastructure competition in many fixed telecommunications markets challenges the traditional legitimation for wholesale regulation in
4
1.1 Research questions
principle as the bottleneck resource is in fact no longer essential (Renda, 2010). Moreover, theoretical studies show that wholesale competition in the case of two distinct
infrastructures can lead to competitive outcomes in wholesale and retail markets, although monopoly-like outcomes may emerge in specific cases (Bourreau et al., 2011).
Therefore, the presence of duopolistic wholesale competition may be sufficient to ensure OA. However, empirical studies show that tacit collusion, which is so far not considered in the theoretical studies on wholesale competition, can significantly alter market outcomes in oligopolies with few competitors. Thus, a controlled empirical evaluation of the effects of wholesale competition is warranted to test whether it can effectively sustain competitive retail markets. Moreover, this calls for a theoretical analysis
of vertically integrated firms’ incentives to tacitly coordinate in upstream and downstream markets (Nocke and White, 2007; Normann, 2009). With regard to these issues
the third research question considers:
R ESEARCH Q UESTION 3. Given infrastructure competition, is wholesale competition sufficient to ensure Open Access for downstream retailers? Does wholesale competition benefit
consumers?
In the context of recent merger cases in the mobile telecommunications industry, competition authorities have repeatedly been confronted with the decision whether potential efficiency gains, due to consolidation, outweigh diminished competitiveness, due
to a lower number of firms in the market. More specifically, there is the concern that
implicit cooperation among competitors in the form of tacit collusion may be facilitated
with fewer firms and thus competitiveness could be aggravated over and beyond the
concentration of market power. Thus, an evaluation of anti-competitive effects seems
of particular relevance if market concentration is reduced from four to three or even
from three to two competitors. However, the detection of tacit collusion based on field
data is generally difficult, because precise information on the cost structure is often
missing. Yet, it is even more difficult in merger control proceedings where hypothetical counterfactuals have to be estimated and assessed ex ante. Therefore, economic
laboratory experiments are well suited to empirically examine the systematic effects on
5
Chapter 1 Introduction
tacit collusion in a controlled environment. Although, it is often assumed that there is
a strictly monotonically decreasing relationship between the number of firms and the
competitiveness of a market (see, e.g., Potters and Suetens, 2013), a systematic evaluation of number effects that also controls for different competition models is missing
so far. Moreover, the studies which examine number effects in more specific contexts
(see, e.g., Huck et al., 2004b, for homogeneous Cournot competition) do not distinguish
between behavioral effects, i.e., tacit collusion, and structural effects, i.e., equilibrium
predictions. Finally, there is no experimental analysis that extends the examination
of number effects in the context of symmetric firms to markets with asymmetric distribution of market power. Yet, this scenario is particularly relevant in cases where
incumbent operators remain in a dominant position, but competing operators have
established significant market shares such that regulatory agencies consider (partial)
deregulation (see, e.g., Ofcom, 2014; Bundesnetzagentur, 2015, for discussions of number effects in a regulatory context). In essence, the fourth research question reads:
R ESEARCH Q UESTION 4. What is the relationship between the number of competitors and
the competitiveness of an oligopolistic market?
Economic laboratory experiments concerned with competitive settings and firms’
strategies have traditionally been conducted in discrete time, i.e., either as a
simultaneous-move or a sequential-move game. However, fixing the timing of decisions according to a pre-specified order abstracts from an important dimension of
strategic decision making in practice: the decision on the timing of an action. Recently, studies have employed experimental designs with a continuous time framework, among others, in the context of a prisoner’s dilemma game (Bigoni et al., 2015;
Friedman and Oprea, 2012), in a network formation game (Berninghaus et al., 2006,
2007) and in a Hotelling (1929) location model (Kephart and Friedman, 2015; Kephart
and Rose, 2015). A continuous time framework endogenizes timing decisions in the
laboratory experiments and thus extends subjects’ action space by allowing for asynchronous interactions. Experimental design in continuous time may be deemed more
realistic for particular scenarios and thus may increase external validity of experimental
6
1.1 Research questions
results. Moreover, continuous time may encourage exploration of the action space as
restrictions on the number of decisions are lifted and also allows for shorter feedback
cycles. Yet, the effects of a continuous time framework relative to discrete time have
not been investigated in the context of oligopoly competition. The following research
question addresses these methodological considerations:
R ESEARCH Q UESTION 5. Does continuous time in experiments on oligopoly competition facilitate tacit collusion relative to discrete time?
In digital services markets, the upstream resource to which (open) access may be
granted is represented rather by virtual goods, such as data, intellectual property or algorithms than by physical infrastructure (see, e.g., Easley et al., 2015; Krämer and Wohlfarth, 2015). Current antitrust cases that center around online intermediaries and their
alleged market power on the basis of unique data sets and the ensuing information advantage exemplify the value that is attributed to these resources (Argenton and Prüfer,
2012). Yet, there also exist voluntary access relationships and agreements among competitors (Mantena and Saha, 2012; Mantovani and Ruiz-Aliseda, 2015). In particular,
social logins, such as “Log in with Facebook”, enable the respective social network and
the content providers to share data, which individually improves their ability to place
targeted advertising. Whereas these mechanisms can improve a website’s user experience and therefore enjoy great popularity among content providers and users alike
in practice, it is a priori unclear how these collaborative efforts impact the competition
between content providers. In particular, data sharing may affect competition between
special-interest content providers from the same domain with regard to users’ choice
to visit the respective platform but also between a general-interest content provider
(the social network) and those content providers with regard to the display of advertising. Therefore, a theoretical microfoundation that relates the effects of data collection
and the ability to target users to ensuing advertising profits in the context of competing outlets is required. Moreover, with regard to (data) neutrality considerations it is
unknown whether voluntary agreements are sufficient to ultimately maximize users’
benefits. The sixth research question summarizes these issues as follows:
7
Chapter 1 Introduction
R ESEARCH Q UESTION 6. What are the competitive effects of social logins that allow for reciprocal access to user and usage data among online content providers?
1.2 Structure of this thesis
The remainder of this thesis addresses the above research questions and is organized
as follows. Chapter 2 provides an introduction to the concept of Open Access in the
context of the current transition to next-generation access networks at the telecommunications infrastructure layer.1 Addressing Research Question 1, the chapter attempts
to reconcile the diverse views on OA by offering a definition and a conceptual framework by which OA endeavors can be identified and uniquely classified. Along this
framework, the extant economic literature is surveyed with regard to aspects of competition and social welfare, investment and innovation, as well as practical and legal
issues. Based on these insights, a policy guideline is developed that may assist policy
makers in identifying the appropriate OA scenario for the regulation of telecommunications infrastructure.
The methodological foundations of this thesis are laid out in Chapter 3.2 Based on
the proposal of an idealized research process cycle for microeconomically founded IS
research, it is argued how theoretical and empirical research approaches may be employed to develop and refine robust theory. Building on this framework, it is argued
how analytic models and laboratory experiments can be combined to serve as a testbed
for regulatory design proposals. Finally, Research Question 5 is investigated by an
experimental study that compares oligopolistic competition under continuous and discrete time.
Chapter 4 addresses Research Question 2 and considers a game-theoretic model of
margin squeeze regulation in a market with horizontally differentiated competition be1 Chapter
2 is based on joint work with Jan Krämer (Krämer and Schnurr, 2014).
3.1 is based on joint work with Jan Krämer (Krämer and Schnurr, 2016). Section 3.3 is based on
joint work with Niklas Horstmann and Jan Krämer (Horstmann et al., 2015).
2 Section
8
1.2 Structure of this thesis
tween two vertically integrated firms and one non-integrated retailer, whereby the nonintegrated firm relies on wholesale access provided by one of the integrated firms.3
On the basis of the methodological foundations set out in Chapter 3 and the theoretical
analysis in Chapter 4, Research Questions 2 and 3 are examined empirically in Chapter 5.4 More specifically, Chapter 5 employs a continuous time economic laboratory experiment with both student and expert participants to compare market outcomes under
different modes of wholesale competition as well as under an OA regulation preventing a margin squeeze. A theoretical explanation for the obtained experimental results
is provided based on a comparative analysis of collusion incentives among vertically
integrated firms in the cases of a wholesale monopoly and wholesale competition.
Next, Chapter 6 conducts a meta-analysis of the literature on oligopoly experiments
and two economic laboratory experiments to address Research Question 4.5 Therefore,
the number of firms as well as the mode of competition are systematically varied in an
experimental analysis of symmetric firms. Furthermore, number effects between three
and four firms are examined for asymmetric distributions of market power.
Chapter 7 focuses on voluntary access relationships at the services layer of the ICT
value chain and addresses Research Question 6.6 Therefore, content providers’ strategic decisions whether to offer and/or adopt a social login are scrutinized and feasible
market outcomes are identified. Thus, the incentives and fundamental trade-offs that
drive adoption decisions and impact the profitability of content providers are examined
based on a stylized model of horizontal competition between special-interest content
providers, and advertising competition between special-interest content providers and
the general-interest content providers.
Finally, Chapter 8 concludes and discusses limitations of this thesis together with
promising avenues for future research.
3 Chapter
4 is based on joint work with Niklas Horstmann and Jan Krämer (Horstmann et al., 2016a).
5 is based on joint work with Niklas Horstmann and Jan Krämer (Horstmann et al., 2016c).
5 Chapter 6 is based on joint work with Niklas Horstmann and Jan Krämer (Horstmann et al., 2016b).
6 Chapter 7 is based on joint work with Jan Krämer and Michael Wohlfarth (Krämer et al., 2016).
4 Chapter
9
Chapter 2
A Unified Framework for Open Access
Regulation
W
ITH the Digital Agenda 2020 the European Commission has set ambitious
targets for its member states and the European telecommunications indus-
try. The requirements stipulate that until 2020 every household in the European Union
(EU) should be covered by a broadband connection offering at least 30 Mbit/s of bandwidth. Moreover, a penetration rate of above 50% is envisioned for 100 Mbit/s connections. In contrast to the ambitious political goals, the implementation status is far
behind schedule. For instance, by mid 2013 only 2% of European households have already subscribed to a connection offering 100 Mbit/s or more (European Commission,
2013b). Thus, large investments are needed to upgrade the existing broadband networks to the desired level. Especially, the deployment of NGANs represents the most
substantial share of these investments.
At the same time, European network operators experience declining revenues and profits facing strong competition by alternative infrastructures and Internet Protocol (IP)based services. In particular, former incumbent operators have criticized the current
regulatory regime as heavy-handed and hostile to any investment strategy, portraying
This chapter is based on joint work with Jan Krämer (Krämer and Schnurr, 2014).
11
Chapter 2 A Unified Framework for Open Access Regulation
the European regulatory framework as the underlying root cause for the industry’s bad
performance (ETNO, 2013).
In July 2012, Commissioner Neelie Kroes announced her plan to enhance the broadband investment environment signaling her willingness to lighten the regulatory burden. The industry and various analysts have viewed this promise as a paradigm shift
of the European Commission’s stance towards access regulation. While it will likely
not lead to a complete withdrawal of the regulatory framework, the balance between
static and dynamic efficiency goals is going to be readjusted. Addressing the slow uptake of next-generation networks in Europe compared to Asian countries and the US,
Kroes declared establishing an investment-friendly environment as the primary goal.
While only a year before, the Commission postulated strict unbundling rules based
on cost-based pricing, Kroes now advocated in favor of an approach based on nondiscrimination rules. In addition, she promised to abstain from further price cuts of
wholesale access charges to legacy copper networks and to establish a harmonized stable price floor across Europe (Kroes, 2012).
The first action of the European Commission in light of this announcement is the recommendation on non-discrimination and costing methodologies, which was published in
September 2013. The recommendation outlines the conditions that would allow European regulators to replace cost-based price regulation with non-discrimination obligations, even in the presence of significant market power (European Commission, 2013a).
During the consultation process that followed the draft recommendation, discussions
have evolved around the implementation of non-discrimination. Most of all, contrary
views have been stated on what actually defines a level playing field between the incumbent’s subsidiary and competitors, besides a uniform wholesale price. In particular, network operators oppose an equivalence of input regime, that prescribes equality in
terms of the used infrastructure and processes. Instead they argue in favor of equivalence of output that abstracts from the actual infrastructure and is concerned with equal
functionality. Thus, the debate illustrates the difficulties and conflicts that are hidden
behind the intuitive notion of non-discrimination.
12
Previously, in Europe the idea of non-discriminatory access has been discussed under
the notion of Open Access and in the context of public-sector participation (European
Commission, 2009b). It has frequently been stated that OA could provide a balance
between static and dynamic efficiency (e.g., Klumpp and Su, 2010; OECD, 2013). Yet,
OA has been used to describe a very diverse set of access concepts. While there is no
explicit definition given by regulators or legislators, the term has been used in various
contexts of access regulation, state aid and voluntary provision of wholesale access
provision. Despite the widespread use, there is no common understanding of the term
among scholars, regulators and industry practitioners. Therefore, a clarification of the
actual OA notion and the related concept of non-discrimination is needed. In particular,
a structured evaluation of the diverse applications is required in order to allow for
precise policy conclusions that can guide the search for a new European regulatory
framework.
With regard to this ongoing discussion, this chapter is concerned with the application of
OA at the network infrastructure level as well as with current regulatory issues and use
cases that have influenced the European debate. At the same time, the history of OA
as a regulatory remedy goes back for several decades and encompasses applications in
telecommunications, but also in other industries such as the media sector. Policy debates about appropriate access provisions within the US have coined and significantly
shaped the understanding of the OA principle. While covering the details of these historic applications is beyond the scope of this study, information drawn from the US
is included when it can be applied to and interpreted in the European NGAN context.
Moreover, this chapter does not explicitly address the net neutrality controversy (which
is, e.g., surveyed by Krämer et al., 2013) nor a comparison of both concepts (which is,
e.g., discussed by Hogendorn, 2007). However, the proposed framework may serve as
the basis for further refinements and extensions that focus particularly on quality of
service (QoS) characteristics and requirements in access relationships among network
operators as well as between network operators and application services providers. On
top of the telecommunications network infrastructure, digital convergence is likely to
raise new questions whether traditional network concepts should be applied to higher
13
Chapter 2 A Unified Framework for Open Access Regulation
layers of the value chain. Therefore, the discussion at the end of this chapter points to
potential applications of OA at the services level of the ICT value chain.
Along these lines, the remainder of this chapter is structured as follows: In Section 2.1
the various notions of OA that were proposed by different stakeholders are presented
and subsequently reconciled into a unified definition. Moreover, a conceptual framework is developed that allows for the classification of the diverse OA application scenarios. Based on this classification, in Section 2.2 the extant economic literature is reviewed and policy implications are derived for each OA application scenario. Section 2.3 relates the various OA applications to each other and presents an overreaching
policy guideline for the OA regulation of NGAN. Finally, Section 2.4 discusses the main
results and identifies possible limitations and extensions.
2.1 The concept of Open Access
There is a fundamental lack in common understanding what actually defines an OA
policy and along which dimensions OA regulation can be structured. For example,
while OA has been used to describe access obligations including price regulation in
the US (Speta, 2000; Farrell and Weiser, 2003), the European Commission’s understanding of OA refers to mandated access in the case of state aid (European Commission,
2013c), and on the other end network operators have put emphasis on voluntary access
(Deutsche Telekom, 2011).
In the following, the definitions of OA proposed by the European Commission, the German telecommunications industry, and proponents of the open access network model
are presented. The definitions indicate that there is common ground in referring to
non-discrimination as the central criterion, but they also illustrate that stakeholders
highlight different additional aspects. As mentioned above, these aspects differ with
respect to how open access terms shall be reached (mandated or voluntary), but also
with respect to which access levels of the value chain are concerned and whether the
14
2.1 The concept of Open Access
notion of OA requires a specific organizational form of the access provider (vertical
separation, public-sector participation).
2.1.1 Open Access in the context of public-sector participation
The emphasis on the European context is founded in the European Commission’s use
of the actual term in the State Aid Guidelines in 2009, thus introducing OA as a legal
criterion that European network operators have to fulfill when they receive state aid
(European Commission, 2009b). Since the European Commission has refrained from
giving an explicit legal definition, various stakeholders have subsequently engaged in
interpretations and new definitions of the term, in particular in the context of publicsector participation, but also with regard to a potentially larger application scope.
The implicit definition that can be derived from the State Aid Guidelines is summarized by the Body of European Regulators for Electronic Communications (BEREC,
2011, p.8):
“The term ‘open access’ [...] refers to mandated wholesale access whereby
operators are offered effective, transparent and non-discriminatory
wholesale-access to the subsidized network(s).”
The notion of non-discrimination is defined in Article 10 of the European Access Directive and requires equality between the integrated downstream subsidiary and an independent retailer as well as between two independent retailers (European Commission,
2002). However, this does not prohibit differentiated access offers in general. In particular, different prices can be charged if this differentiation is based on objectively justifiable reasons. The recommendation on non-discrimination explicitly mentions that
volume discounts and/or long-term access pricing can be compatible with the nondiscrimination criterion (European Commission, 2013a). In order to safeguard effective
implementation in the context of such volume discounts, the recommendation stipu-
15
Chapter 2 A Unified Framework for Open Access Regulation
lates that favorable conditions to the subsidiary are not allowed to exceed the highest
discount offered to independent downstream firms.
In the US context, OA has been mentioned, on the one hand, in association with stateowned municipal networks (Lehr et al., 2004), where OA was seen as an instrument to
ensure non-discriminatory access and may be adopted voluntarily or mandated. On
the other hand, OA has frequently been used interchangeably with mandated priceregulation (Speta, 2000).
2.1.2 Voluntary Open Access
In Germany, OA has played a prominent role in the debate about potential regulatory
regimes governing NGANs. In the context of regional deployment of NGA networks
and in the advent of new business models, OA has been seen as a solution to drive rapid
construction, fast penetration and interoperability. The Federal Network Agency, Germany’s national regulatory authority (NRA), established the “NGA-Forum" with the
goal to promote the standardization of access products which it views as a prerequisite
for symmetric OA. In contrast to the European Commission, which has characterized
OA as mandated wholesale access, network operators here have tried to establish OA
as a concept that relies on the voluntary decision by the particular access provider. A
definition representing the consensus among network operators was presented at the
German IT-Summit (2010, p.4):
“Open Access in FTTB/FTTH-Networks refers to the voluntary, nondiscriminatory access at different levels of the value chain.”
2.1.3 The Open Access network model
Already prior to the State Aid Guidelines, a strand of literature emerged based on a
notion of OA which not only requires non-discrimination, but also functional separa-
16
2.1 The concept of Open Access
tion between network operators and the services companies. The OA network model
is hereby characterized by the separation of roles between the service provider and the
network owner. Battiti et al. (2005) stipulate the belief that vertical integration of communication networks is the main reason for “high costs of services and barriers of competition" (p.1). While a further economic analysis or elaborated reasoning to back the
hypothesis of hindered innovation due to vertical integration is often neglected, proponents instead point to public infrastructure such as roads in the case of transportation
as a proven benchmark. In their view, a concept relying on mutual control and shared
usage of physical access networks is able to lower costs for deployment and usage of access networks, while providing users with a greater choice and service providers with
more freedom (Battiti et al., 2005). This line of argument focuses rather on aspects of
static efficiency of network operations, e.g., fair competition among service providers,
than aspects of dynamic efficiency as actual investment for building networks is believed to be facilitated by technological innovation, e.g., wireless networks (Bogliolo,
2009). A summarizing definition of OA as viewed by this strand of the literature is
given by Forzati et al. (2010, p.1):
“In the open access network model, the roles of the service provider and
the network owner are separated, and the service providers get access to
network and the end customers on fair and non-discriminatory conditions.”
As can be seen by the manifold definitions, the criterion of non-discrimination is central
to the concept of OA. However, there is a diverse understanding among stakeholders of
how the goal of non-discriminatory access is effectively and efficiently achieved. Therefore, the definitions differ according to the requirements they postulate and whether
OA should be established as a voluntary or regulated regime.
Based on these insights the following unifying definition is proposed:
Definition 2.1 (Open Access Regulation). Open Access regulation refers to the mandated
or voluntary provision of access to an upstream resource which must be based on the principle
of non-discrimination. The concept may apply to publicly or privately owned access providers
17
Chapter 2 A Unified Framework for Open Access Regulation
that are vertically separated, integrated or represent a cooperative of multiple entities. Open
Access regulation usually refers to the network layer, but may also be applied to other layers of
the telecommunications value chain.
Considering this definition, it becomes clear that non-discrimination may be achieved
by various approaches. In fact, even price-regulation can be included as a specific version of mandated OA that attempts to implement non-discrimination by setting a regulated access charge. In the narrow sense, however, Open Access refers to mandated
non-discrimination, where the upstream provider may freely set the terms of access, but
is forced to provide access on a non-discriminatory basis. Thus, following the previous discussion, the proposed definition also reconciles the different meanings of OA
in Europe and the US. As a consequence, scholars analyzing OA have to be aware of
the large scope that the concept actually comprises and derived implications need to be
explicitly related to a particular application of OA.
2.1.4 A conceptual Open Access framework
In order to allow for the classification of results obtained by the extant economic literature and in order to be able to derive coherent policy conclusions, a conceptual framework is offered that structures the further analysis of particular OA relationships and
allows for a subsequent comparison of different OA models. The OA classification
framework (Figure 2.1) is based on three dimensions that are deduced from the key
characteristics of the presented definitions and represent the major determinants of access relationships in the NGAN context.
The vertical structure denotes how ownership in the access network (the upstream market) and activities in the services market (the downstream market) are related. Of
course, the actual degree of integration may not only be defined by common ownership, but also by additional dimensions such as task integration, knowledge integration and coordination integration (Jaspers and van den Ende, 2006). While being aware
of the fine-granular spectrum of vertical organization models, the subsequent analysis
18
2.1 The concept of Open Access
is structured by referring to the most relevant cases, i.e., i) an integrated firm that is
also active in the downstream market, ii) a separated wholesaler, or iii) a cooperative
undertaking between several downstream competitors (co-investment).
Ownership denotes the ownership structure and the goals of the access provider that
vary with the influence of the public sector. The access provider may be entirely stateowned, as in the case of municipal public utilities or public access networks (e.g.,
the Australian National Broadband Network), represent a public-private-partnership
(PPP) or a private-sector, profit-oriented corporation. PPP models again can be differentiated according to the allocation of responsibilities concerning financing, design, construction and operation of the access network (European Commission, 2011). The goals
of the organization according to its ownership structure may range from pure profitmaximization to non-profit pursuit of public interests. A further differentiation may be
needed in the case of a private network operator specifying whether the firm is subject
to price-regulation or whether it is unconstrained in setting its wholesale prices.
The access level, finally, indicates at which level of the value chain access is given to
downstream competitors. In this vein, the concept is readily applicable to a wide and
diverse set of access relationships. Since the emphasis of this article is on NGANs, the
following analysis is particularly concerned with access levels to “Internet infrastructure services", i.e., Open Systems Interconnection (OSI) layers one through three, according to the terminology introduced by Jordan (2009). While most incentive-based arguments concerning vertical integration and public ownership apply equally to higher
levels of the Internet value chain, specific characteristics of “Internet application services" (Jordan, 2009) need to be considered in addition.1 The set of feasible access options to network facilities may differ across access technologies (e.g., coaxial, copper,
fiber networks) and depend as well on the network architecture (e.g., point-to-point
and point-to-multipoint). Access may also be granted at different geographical locations. Moreover, the introduction of QoS and respective traffic classes may increase the
technical and commercial possibilities of further differentiation in access agreements
1 Section
2.4 discusses how the framework can also be applied to higher access levels. Chapter 7 presents
an analysis of voluntary (open) access to user and usage data in online content markets.
19
Chapter 2 A Unified Framework for Open Access Regulation
between network operators. Eventually, the access level defines the degree of control
and the potential quality differentiation that the access seeker can achieve. Therefore,
this dimension reflects also a well-known concept in telecommunications regulation:
the Ladder of Investment (Cave and Vogelsang, 2003; Cave, 2006a). Figure 2.1 illustrates
potential access levels in the case of a fixed NGAN. A more elaborate discussion of the
Internet layers as well as architectural principles can be found in van Schewick (2010).
Claffy and Clark (2013) point to a further important differentiation when considering
access to (infrastructure) platforms in convergent networks, namely the distinction of
the platform’s intended use (closed vs. open to complementors) and its construction
(single- vs. multi-firm formation). Thus, their framework is especially suited to guide
consistent regulation of emerging specialized or managed services and coexisting open
systems like the Internet that are based on identical physical networks.
F IGURE 2.1: OA classification framework: vertical structure (impact on downstream market),
ownership (goal of organization), and access level (quality differentiation).
Vertical Structure
Vertical
Integration
Ownership
Right of Way
Investment
Cooperative
Ducts
Physical Unbundling
Bitstream Access
Vertical
Separation
Public
Ownership
Resale
PPP
Private
Ownership
Access Level
Based on these three dimensions bilateral access relationship can be characterized, but
also more complex ventures (as, e.g., proposed by Forzati et al., 2010) can be illus-
20
2.1 The concept of Open Access
trated by displaying the distinct and potentially heterogeneous individual relationships. Note, however, that access relationships are primarily defined by the vertical
structure and the ownership, whereas the access level defines the spectrum of this particular relationship. Therefore, it is sufficient to consider the dimensions of vertical
structure and ownership to distinguish four general settings of potential OA applications (as illustrated by Figure 2.2).
1. Vertically integrated network and services providers that are typically represented by national incumbents or regional operators of NGANs. In this context,
an obligation to provide non-discriminatory access may be seen as an alternative instrument to cost-based price regulation. Further discussions have evolved
around the question whether OA could be realized as a voluntary concept.
2. Vertically separated, profit-maximizing network operators including organizations that were established by vertical separation of formerly integrated operators
(e.g., Openreach in the UK ) and cross-industry entry of companies that are active
in other markets such as energy utilities (e.g., RWE Germany). OA is envisioned
to stimulate wholesale agreements by means of transparency and standardization
resulting as a consequence of the non-discrimination condition.
3. Cooperative undertakings by private-sector organizations including risk- and
network-sharing contracts as well as agreements on geographically complementary investments. Here, OA may serve as a regulatory tool to govern ex-ante
access to the cooperative or ex-post access to the network.
4. Public-sector participation including local and nation-wide initiatives, state aid,
public outsourcing, and public-private joint-ventures where the physical network
is designed, built and maintained by the public entity while private firms operate
the active network facilities and provide services. In this context OA is usually
seen as an instrument to minimize the distortion of competition by state activity
and is in general stipulated as a mandatory obligation.
21
Chapter 2 A Unified Framework for Open Access Regulation
F IGURE 2.2: Classification of OA application scenarios.
Vertical
Structure
Vertical
Integration
4.
Service of general
interest
1.
Mandated nondiscrimination
Price regulation
Investment
Cooperative
Public infrastructure &
private operators (PPPs)
Formation of cooperative
undertakings
3.
Risk-sharing instrument
Vertical
Separation
Nation-wide neutral
access networks
State aid: prerequisite
for financing & subsidization
Local broadband deployment: demand
stimulation & infrastructure utilization
Public
Ownership
PublicPrivatePartnership
Mandated vertical
separation
2.
Market entry of crossindustry players
Private
Ownership
Ownership
2.2 Literature survey
Based on the preceding classification, each of the identified OA applications is assessed
from an economic perspective. To this end, implications for each of the respective forms
of vertical structure and ownership are derived by surveying the extant economic literature. A particular emphasis is put on the presentation of the prevailing effects that
have been identified by scholars as determinants of static and dynamic efficiency. In
addition, regulatory and legal aspects that follow from particular characteristics of the
access provider are noted. More precisely, the following list comprises the set of efficiency measures and determinants that are used throughout the assessment:
Static efficiency is measured by social welfare and constituted by allocative and
productive efficiency. Relevant determinants include the presence of transaction
costs, economies of scale and scope, price and non-price discrimination, foreclosure, intensity of competition, and externalities.
Dynamic efficiency encompasses technological progress and innovation, i.e., improvements in productive efficiency, which can be measured by social welfare
22
2.2 Literature survey
over time (Kolasky and Dick, 2003; Viscusi et al., 2005). The literature on telecommunications has employed several proxies in order to quantify dynamic efficiency, such as the magnitude of investments, coverage, innovation, or the extent of infrastructure-based competition. In vertically related industries these outcomes are affected by the degree of coordination between upstream and downstream segment, and by the prevailing investment incentives of the incumbent as
well as potential entrants.
Regulatory and legal requirements describe the necessary degree of information and
monitoring capability, accountability, task complexity and effectiveness.
2.2.1 Vertical integration and separation of the access provider
Liberalization in the telecommunications sector was driven by the perception that telephony service and long-distance networks could potentially be served by multiple competing firms. In contrast, the access network was still seen as a natural monopoly by the
consensus opinion. Accordingly, questions concerning vertical separation have mostly
been analyzed by assuming a monopolistic supplier that serves an oligopolistic or competitive downstream market. Addressing doubts about the natural monopoly assumption due to new access technologies (mobile, fixed-wireless), the recent literature on
investment and regulation (see, e.g., Vareda, 2011) and on price discrimination (Inderst
and Valletti, 2009) has extended the conventional model by allowing for potential replication of the upstream resource through market entry.
In general, the literature can further be distinguished according to whether the upstream firm possesses freedom in setting wholesale prices or if prices are set by the regulator. Particularly, the impact of price regulation on investment incentives has been
thoroughly analyzed by a recent strand of literature. The results and remaining research questions of this literature are covered in a survey by Cambini and Jiang (2009),
and therefore, they are not included in this overview. Further summaries that deal
with specific aspects of the vertical structure in the telecommunications industry can
23
Chapter 2 A Unified Framework for Open Access Regulation
be found in Tropina et al. (2010), Janssen and Mendys-Kamphorst (2008), Gonçalves
and Nascimento (2010), and Yoo (2002).
Riordan (2008) presents a summary of the major theories explaining benefits and weaknesses of vertical integration in terms of static efficiency. It is shown that the perception about vertical integration as a beneficial or worrying practice has changed repeatedly since the 1950s according to the predominant theory at a particular epoche. Major insights have been obtained by the structure-conduct-performance theory dealing
with vertical foreclosure and leverage of market power, the Chicago School stipulating the theories of single monopoly profit and elimination of price markups (Spengler, 1950), Transaction Cost Economics developing the theory of incomplete contracts
(Whinston, 2003), and Post-Chicago Economics by pointing out to incentives on restoring monopoly power (Rey and Tirole, 2007b).
The general question what actually determines firms’ boundaries and the degree of vertical integration is adressed in the first part of an extensive review by Lafontaine and
Slade (2007). The authors present the most prominent theories along with prototypical
analytical models and investigate whether derived predictions are supported by empirical findings. First, there is strong empirical support for the hypotheses derived from
Transaction Cost Economics with regard to backward-integration. High asset specificity, transaction complexity and uncertainty are identified as drivers of integration
between suppliers and manufacturers. Second, moral-hazard arguments established
by agency theory seem to explain forward-integration between manufacturers and retailers very well, with the exception of the empirically observed negative relationship
between downstream risk and vertical integration. The second part of the survey summarizes the observations made by the empirical literature which has studied the effects
of vertical integration on firms’ performance and on consumer surplus. The obtained
conclusions are included separately hereafter in order to relate the empirical findings
to the respective underlying theoretical concept.
24
2.2 Literature survey
The Single Monopoly Profit Theorem (SMPT)
The theorem stipulates that an up-
stream monopolist will generally not inefficiently leverage its market power to complementary or downstream markets since it can generate the monopoly profit only once.
Due to the incentive to maximize output of the complementary downstream products,
the monopolist will decide to integrate forward only in the case if the downstream
market is inefficient. However, this conclusion is subject to a number of exceptions as
presented by Farrell and Weiser (2003) and builds on the following assumptions. First,
the monopolist must have the ability to make enforceable multilateral commitments as
shown by the theory of restoring monopoly power (Rey and Tirole, 2007b). Second, the
monopolist must posses sufficient freedom to set wholesale charges in order to extract
the monopoly rent. This is prohibited by most regulatory regimes that implement price
regulation on a cost basis. In fact, it is the goal of regulated industries to prevent the
monopoly profit in the first place. Moreover, the SMPT fails if the upstream monopoly
is under threat itself by potential “two-level entry" and may be protected by foreclosing
the downstream market.
Restoring monopoly power
Rey and Tirole (2007b) show that the ability by a monop-
olistic supplier to fully extract the monopoly profit breaks down if the wholesale firm
is unable to commit to multilateral contracts. The monopolist has an incentive to offer
a lower marginal access price to at least one downstream firm while taking a higher
fixed premium. The downstream firm will accept the deal if the competitive advantage of the lower input price outweighs the additional fixed cost. Since downstream
competitors anticipate this incentive to discriminate, the supplier is unable to set the
monopoly price in the first place. The network infrastructure, therefore, exhibits characteristics of a durable good (Coase, 1972). The commitment problem can be resolved
and monopoly power can be restored by the supplier entering the downstream market through an own subsidiary. Alternatively, the separated monopolist may circumvent the commitment problem by negotiating an exclusive agreement with one of the
downstream firms. Paradoxically, monopoly power may also be restored by regulatory
activity. Rey and Tirole (2007b) show that a non-discrimination criterion (other than a
25
Chapter 2 A Unified Framework for Open Access Regulation
most-favored-customer clause) will prohibit the supplier from providing preferential
treatment to a downstream firm and bind the supplier to the uniform (monopolistic)
access charge.
Raising rivals’ costs (RRC) Critics of vertical integration have frequently raised the
concern that a monopolistic supplier that is active in the upstream and downstream
market has an incentive to distort downstream competition by raising rivals’ costs. An
assessment of this anti-competitive behavior has first been conducted by the seminal
work of Salop and Scheffman (1983, 1987) and was followed by extensive analysis of
RRC in the context of specific characteristics of network industries. In general, the
integrated firm is able to raise costs either directly via the access charge (price discrimination) or through sabotage in the process of providing the input good (non-price
discrimination). The former approach differs from the latter by the additional direct
income effect for the upstream firm due to a higher wholesale price. In both cases the
integrated firm may benefit from a competitive advantage in the downstream market
derived from lower marginal costs. At the same time, decreasing market shares of independent retailers will lead to a diminished wholesale profit. Therefore, the decision
to raise rivals’ costs is always associated with a trade-off between upstream and downstream profits. In general, the empirical evidence with respect to observed foreclosure
and RRC behavior according to Lafontaine and Slade (2007) is mixed, but there are
several studies that uncover evidence of foreclosure in the case of industries characterized by natural monopolies, most notably in Cable TV. However, several studies point
to the fact that efficiency gains due to vertical integration may outweigh the costs of
foreclosure and could therefore benefit consumers through lower prices.
Price discrimination
Assessing the incentives and welfare implications of price dis-
crimination in the case of a regulated upstream market, Vickers (1995) shows that the
integrated firm has an anti-competitive incentive to favor its subsidiary. Since entry
in the downstream market is assumed to be costly in terms of duplication costs, it is
concluded that price discrimination may also have a positive effect on total welfare by
26
2.2 Literature survey
reducing inefficient entry. Likewise, Reitzes and Woroch (2009) find that the integrated
input monopolist will engage in price discrimination against downstream rivals. However, the integrated monopolist may also set an input price above marginal cost to its
downstream affiliate.
For an unregulated industry, the ability to discriminate in wholesale prices leads to contrary results depending on whether upstream firm faces the threat of demand-side substitution. DeGraba (1990) shows that under the assumption of an unconstrained monopolist, the more efficient downstream firm is optimally charged higher input prices.
However, with respect to dynamic efficiency a ban of price discrimination increases investment incentives of retail firms. When introducing the possibility for demand-side
substitution these results are reversed (Inderst and Valletti, 2009). In particular, more
efficient firms receive a discount compared to their less efficient competitors. Thereby,
price discrimination hurts consumers in the short-run, but improves investment incentives, thus creating long-run benefits.
Non-price discrimination The literature on sabotage distinguishes between cost raising strategies that increase rivals’ per-unit costs on top of the wholesale rental charge,
and degradation of quality that lowers the demand of downstream competitors (Brito
et al., 2012). Under a linear demand structure cost-raising and demand-reducing sabotage are indistinguishable (Reitzes and Woroch, 2009). Examples of non-price discrimination include poor quality of interconnection, delay in processing orders, creation
of incompatibility, bundling of complements (Economides, 1998), or preferential treatment of affiliated content (Brito et al., 2012).
Incentives for sabotage emerge only in the presence of binding input price regulation
(Beard et al., 2001; Reitzes and Woroch, 2009) and in the case that downstream firms
can reap profits, i.e., that the downstream market is not perfectly competitive (Mandy,
2000). If prices are regulated, the input monopolist has to weigh higher downstream
profits against lost upstream earnings when deciding whether to engage in non-price
discrimination. While raising rivals’ cost will lead to an increase in downstream prices
27
Chapter 2 A Unified Framework for Open Access Regulation
and higher output by the integrated affiliate, the reduced market share of independents
will reduce the demand for input (Economides, 1998). The theoretical literature has
shown that while in most cases the integrated firm has an incentive to discriminate,
there are exceptions in particular settings.
The results of the many theoretical models developed by scholars differ according to
some key modelling decisions and assumptions about the market structure. As summarized by Mandy (2000), the incentive to discriminate is altered by the downstream
market structure (degree of product differentiation, costs of sabotage to the perpetrator,
presence of scale economies, relative efficiency of competitors, competition intensity),
characteristics of the upstream segment (upstream margin, potential upstream competition), and the model of the integrated firm (separated or common profit maximization).
Some key results are highlighted in the following.
Economides (1998) stipulates that the sum of the three described income effects incurred by non-price discrimination is generally positive, and that the monopolist raises
costs until competitors are forced out of business. According to this model, the result
remains unaffected by a cost advantage or disadvantage of the monopolist’s subsidiary
compared to independent downstream firms.
Mandy (2000) concludes that the integrated monopolist abstains from sabotage if a) the
downstream market is competitive (low degree of double marginalization), b) there is
a sufficient upstream profit margin, and c) the monopolist’s subsidiary is substantially
less efficient compared to its competitors. Referring to the latter aspect, Weisman and
Kang (2001) highlight that the incentive to discriminate is most pronounced in the situation where integration is highly efficient. Thus, the regulator is presented with the
challenge to choose between productive and allocative efficiency. Symmetric considerations hold for the upstream market: If the monopolist would be required to decrease its
market share, wholesale revenues decrease and the incentive to discriminate is raised
(Weisman, 1995).
28
2.2 Literature survey
In the case that the upstream monopolist integrates forward and represents a new entrant into the downstream market, Sibley and Weisman (1998) stipulate that for initially
low market shares, the integrated firm has no incentive to raise rivals’ costs. Analyzing the impact of downstream competition, Mandy and Sappington (2007) confirm the
stated results for cost raising strategies in quantity-setting (Cournot) and price-setting
(Betrand) competition, as well as cost raising sabotage in quantity-setting competition,
but find opposing results for quality reducing behavior in price-setting competition.
Brito et al. (2012) analyze quality degradation in the presence of product differentiation.
The authors show that the integrated firm has an incentive to discriminate, if the competitor provides an inferior product compared to its own subsidiary and the upstream
margin is sufficiently low. They further show that a disparity in quality and low access
prices may also lead to discrimination in the separation case, since the upstream firm
benefits from a larger market share of firms that produce superior quality products.
With regard to regulated upstream prices, the integrated firm has an unambiguous incentive to discriminate if the access charge is set equal to marginal costs of providing the
input (Sibley and Weisman, 1998; Kondaurova and Weisman, 2003; Reitzes and Woroch,
2009). Obviously, the monopolist will engage in sabotage since there are no opportunity costs in terms of lost wholesale profits. Only in the case of price competition this
result may be reversed (Mandy and Sappington, 2007).
Reitzes and Woroch (2009) highlight the fact that pricing parity in the form of cost-based
regulation incentivizes the upstream firm to provide excessive quality to its integrated
affiliate, allowing it to charge higher access fees to its downstream competitors while
degrading their input quality.
Double marginalization A well known beneficial effect of integration in terms of consumer and total welfare is the elimination of markups that may exist in a vertically
separated industry. Except for the case of a perfectly competitive downstream market, double marginalization leads to the contraction of output and higher retail prices
29
Chapter 2 A Unified Framework for Open Access Regulation
if price discrimination is difficult. In contrast, if the upstream monopolist is able to discriminate, e.g., by two-part tariffs, no double marginalization occurs, but at the costs of
full monopoly profits. Under integration, i.e., common ownership, the subsidiary will
consider the actual upstream production costs instead of the access price as marginal
costs (Brito et al., 2012). Thus, double marginalization is also prevented in the case of
linear tariffs (Riordan, 2008).
Economies of scale and scope Proponents of vertical integration point out that separation prohibits the exploitation of substantial economies of scale and scope. In fact,
empirical analyses in adjacent network industries have obtained mixed results. While
other utility industries may have rather different technological characteristics and the
magnitude of efficiencies may depend on the historical development of the sector, they
share the necessity of high sunk upfront investment costs in the upstream market that
have to be geared to the needs of the downstream services. In a conceptual comparison
of the vertical structure in infrastructure sectors Pittman (2003) discusses the respective magnitude of vertical economies, but concludes that rapid technological change in
the telecommunications sector complicates the formulation of a universal hypothesis
about the extent of these efficiencies. Thus, in the light of the following ambivalent
empirical results obtained for other infrastructure industries, one should abstain from
a per se hypothesis that vertical economies are exceedingly high in general. Kwoka
(2002), finds evidence for economies of scope in the US electricity sector. In particular,
economies of coordination between generation and distribution are identified as the
main drivers of integration. These findings are in accordance with the early work of
Kaserman and Mayo (1991) and similar results were replicated for the Spanish electricity market (Jara-Dıaz et al., 2004). However, it is further shown by Kwoka that there are
potential instruments besides integration to achieve effective coordination. In particular, holding companies above a certain size can adequately substitute the coordination
function of the integrated firm. In contrast to these studies, Nemoto and Goto (2004)
find almost no economies of vertical integration for a set of Japanese utilities, while
Garcia et al. (2007) show that no significant technological integration economies can be
30
2.2 Literature survey
observed in a sample of the US water industry. Lafontaine and Slade (2007) report on
the results of four studies that have investigated the effect of mandated separation in
the gasoline refining and sales industry. They find an unambiguous negative impact
on consumer surplus through higher costs and prices. Based on a conceptual analysis, De Bijl (2005) argues that separated operators face higher financing costs due to
a lower scale compared to the integration scenario. Crandall and Sidak (2002) argue
that mandated separation will create arbitrary and artificial boundaries that lead to
inherently inefficient organizations. Furthermore, they refer to the indivisibility of intangible assets such as customer loyalty and goodwill. Additional costs are induced
when separation is imposed on an initially integrated firm. Functional as well as ownership separation require costly structural reorganizations and the redesign of business
processes (Cave, 2006b).
Dynamic incentives
Several scholars have argued that a lack of coordination has a
negative impact on the efficiency of investments (De Bijl, 2005; Crandall and Sidak,
2002; Tropina et al., 2010). In a more differentiated analysis, Iossa and Stroffolini (2012)
show that integration is particularly beneficial when little demand information is available, infrastructure cost is low, or investment is highly risky. Moreover, Crandall et al.
(2010) has stated that separation creates a hold-up problem due to the inability to specify complete contracts. In contrast, Cadman (2010) argues that the hold-up problem is
resolved by a sufficiently competitive downstream market.
Farrell and Katz (2000) have shown that integrated firms may have an incentive to
engage in an “investment squeeze", forcing downstream competitors to engage in excessive investments. In fact, investment incentives in the downstream market may be
strengthened by vertical separation. Since firms expect less discriminatory behavior in
this case, perceived uncertainty is reduced and relationship-specific investment may be
increased (Cadman, 2010).
With respect to the incentive to duplicate infrastructure, the consensus indicates that
separation will diminish the incentives for future entry in the upstream market (De Bijl,
31
Chapter 2 A Unified Framework for Open Access Regulation
2005; Tropina et al., 2010). To some degree, this can be seen as the consequence of
mitigated incentives for anti-competitive behavior by the upstream monopolist.
Regulatory and legal aspects
The literature has described several advantages of ver-
tical separation from a regulatory perspective. The break-up into two separate organizations increases transparency, therefore making it significantly easier to oversee
the behavior of the upstream monopolist (Cadman, 2010; Pittman, 2003). Moreover,
the separation of business activities prevents distortion of competition through crosssubsidization. Even in the case of vertical integration, the ability to impose separation
alone can be an effective instrument to motivate the monopolist to adapt its behavior
to regulatory rules (De Bijl, 2005). On the other hand, the decision to impose separation
may be approached with caution. Due to the high costs of separation, the process is
considered to be irreversible (Teppayayon and Bohlin, 2010). Especially in the case of
high uncertainty about the outcome of separation, regulators should therefore abstain
from mandated separation.
Implications Static efficiency effects in integrated telecommunications industries
have been thoroughly examined by the economic literature. Integrated operators are
shown to benefit from economies of scale and scope as well as low information costs
between the retail and wholesale division. The inherent incentive to price-discriminate
has traditionally been countered with regulating wholesale prices, which in turn raises
the issue of non-price discrimination, e.g., by raising rivals’ costs through sabotage. At
the same time the implementation of regulatory control is faced with imperfect information.
OA, in the narrow sense, may represent an opportunity to take a first step towards
deregulation by abstaining from price regulation. Non-discrimination obligations may
facilitate regulatory monitoring by equivalence of input conditions or more easily ob-
32
2.2 Literature survey
TABLE 2.1: Summary - Vertical Integration & Separation.
Description
Int.
Sep.
Description & References
Static efficiency
Double Marginalization
(+)
(–)
Restoring monopoly power
(–)
(0/–)
Price discrimination
(0/–)
(0/–)
Non-price discrimination
(0/–)
(0)
Economies of scale
Economies of scope
(0/+)
(0/+)
(–)
(–)
Separation costs
(0)
(–)
The subsidiary of an integrated firm considers the actual marginal
costs of upstream production (Riordan, 2008; Brito et al., 2012).
Through integration the upstream monopolist can overcome its
commitment problem. (Rey and Tirole, 2007b).
In the absence of regulation, integrated firms may have incentives
to engage in a price squeeze (Vickers, 1995; Farrell and Katz, 2000;
Lafontaine and Slade, 2007). Under separation this anticompetitive incentive is eliminated, however, an uncontested upstream
monopolist will charge more efficient firms a higher price. (DeGraba, 1990; Inderst and Valletti, 2009).
In the case of price-regulation, the integrated operator is likely to
distort downstream competition, but particular exceptions exist
(inter alia: Economides (1998); Mandy (2000); Brito et al. (2012)).
Larger scale decreases financing costs (De Bijl, 2005).
Empirical evidence of scope advantages in the electricity industry (Kwoka, 2002; Jara-Dıaz et al., 2004). Contrary results in the
case of Japanes utilities (Nemoto and Goto, 2004) and the water
industry (Garcia et al., 2007).
Redesign of business processes (Cave, 2006b). Artificial boundaries lead to inefficient organizations (Crandall and Sidak, 2002).
Dynamic efficiency
Coordination of investments
(+)
(0/–)
Hold-up problem
(+)
(0/–)
Investment squeeze
(–)
(0)
Upstream investment incentives
(+/–)
(–)
Downstream investment incentives
(–)
(+)
Regulation
Transparency
(–)
(+)
Coercion
(0)
(+)
Irreversibility
(0)
(–)
A lack of coordination may lead to inefficient investment (De Bijl,
2005; Crandall and Sidak, 2002)
Separation and incomplete contracts lead to opportunistic ex-post
behavior delaying investments (Crandall et al., 2010). Resolved
by a sufficiently competitive downstream market (Cadman, 2010).
Integrated firms may force downstream competitors to engage in
excessive innovation (Farrell and Katz, 2000).
Ambivalent effects under integration (Cambini and Jiang, 2009).
Reduced incentives for infrastructure-based competition in the
case of separation (Tropina et al., 2010; De Bijl, 2005).
Trust induced by separation decreases perceived uncertainty
(Cadman, 2010).
Separation facilitates regulatory oversight (Cadman, 2010;
Pittman, 2003; Economides, 1998) and prevents crosssubsidization (De Bijl, 2005).
The regulator’s ability to impose separation can be used as coercive instrument (De Bijl, 2005).
The decision to separate can not be reversed ex-post, while effects
may be uncertain ex-ante (Teppayayon and Bohlin, 2010).
Note: The signs (+), (0), and (–) qualitatively indicate the consensus of the surveyed literature on whether the issue has
a positive, neutral, or negative impact, respectively, on the efficiency goals or the regulatory and legal requirements.
For instance, the first row indicates that static efficiency is increased (decreased) due to double marginalization under
integration (separation). (+/–) points to contradictory theories or evidence while (0/+) or (0/–) represent intermediate
measures. This notation applies equally to the subsequent tables.
33
Chapter 2 A Unified Framework for Open Access Regulation
tained benchmarks for margin squeeze tests.2 Moreover, imposing OA in the case of an
integrated access provider can potentially increase the transparency of wholesale offers
and prices, therefore lowering transaction costs for wholesale agreements and entry
barriers to service-based competition. However, further research is needed in order to
decide whether particular OA implementations can contain opportunistic behavior by
the integrated operator.
Ensuring OA by imposing separation of upstream from downstream activities represents an effective instrument to counter non-price discrimination by removing the underlying incentives to discriminate against particular downstream firms and by simplifying regulatory oversight. However, the regulator has to be aware that these gains
come at the costs of double marginalization, increased transaction costs, substantial
separation costs, and diseconomies of scale.
The literature on dynamic efficiency points to ambivalent effects with respect to upstream investment incentives. Further research is expected in this area in order to clarify and quantify particular investment incentives. For instance, the coordination of
investments by internalizing incentives of the downstream market is generally stated
as a strong argument in favor of integration. However, this proposition may have to
be reevaluated in the context of NGANs. Due to a layered architecture founded on
the universal IP standard and the perception that fiber technology will not present an
“innovation bottleneck” soon, the negative impact on innovation may not be that substantial in the short term.
2.2.2 Co-investment
Bourreau et al. (2012a) present a brief summary of the advantages and disadvantages
of co-investment. Accordingly, cost-sharing and risk-sharing rules have the potential
to increase NGAN coverage and enhance consumer surplus. They point out that the
2 See Chapter 4 for a theoretical and Chapter 6 for an experimental analysis of margin squeeze tests in the
context of infrastructure and wholesale competition.
34
2.2 Literature survey
frequently mentioned benefits of cost sharing are directly associated with sharing the
subsequent revenues. Therefore, the business case of the particular investment project
does not improve, unless the co-investment approach leads to more product differentiation in the downstream market, subsequently inducing a demand expansion effect.
Nitsche and Wiethaus (2011) characterize risk-sharing as an alternative to regulatory
instruments such as cost-based price-regulation (more specifically long-run incremental costs (LRIC)), fully distributed costs regulation, and regulatory holidays. In the case
that there is no access regulation or risk-sharing agreement, no access to the downstream market is provided to the entrant ex-post. Risk-sharing is found to induce the
strongest degree of competition compared to all other regulatory regimes and to attain
a higher level of investment than the LRIC approach.
Rey and Tirole (2007a) discuss different rules that govern the access to cooperative
undertakings for future entrants. In particular, they compare the models of nondiscriminatory and fully discriminatory cooperatives. While the former allows later
members to join an existent cooperative and use the established infrastructure by paying a cost-based access charge, the latter charges an entry fee in addition to the access
charge. It is shown that investment is discouraged by an “open access" policy since
future entry will marginalize profits to zero. In contrast, investment is encouraged by
a “closed access" model, but at the same time, excessive restriction of access leads to a
suboptimal outcome from a welfare standpoint. Therefore, Rey and Tirole conclude
that the access model involves the essential characteristics of the trade-off between
static and dynamic efficiency (see, e.g., Gayle and Weisman, 2007). In order to improve social welfare, the authors suggest to either constrain the market power of closed
cooperatives or to give open cooperatives an instrument to protect their investment
partially.
Krämer and Vogelsang (2016) investigate collusive behavior in cost-sharing undertakings in an experimental setup. The authors find that co-investment affects retail prices
by two contrary effects with a similar magnitude. On the one hand, co-investment increases prices by facilitating tacit collusion in downstream markets. On the other hand,
35
Chapter 2 A Unified Framework for Open Access Regulation
reduced investment costs due to co-investment are passed on to consumers. Moreover,
the communication between firms, a necessary prerequisite to co-investment, is found
to have a positive impact on investment.
Inderst and Peitz (2012) show that ex-ante access contracts reduce the incentive to duplicate infrastructure, but increase the area that is covered by at least one network compared to investment under ex-post contracts. At the same time, they show that reduced
duplication allows the access provider to engage in price discrimination against independent downstream firms under the general assumption that aggregate demand of
the market is price-dependent. Consumers are therefore harmed by potentially higher
downstream prices, but may benefit from ex-ante contracts making investment profitable in the first place. The authors also mention that ex-ante contracts will naturally
mitigate the hold-up problem, since the ex-ante commitment prohibits ex-post haggling
by the access seeker.
Bourreau et al. (2013) agree that ex-ante cooperative agreements can potentially increase coverage, but qualify these results subject to particular conditions. The authors
show that the results hold only in the presence of a demand-expansion effect, high
cost savings from joint investment or a combination of both. While the former may
be induced by high service differentiation the latter is especially pronounced in the
case of high uncertainty. Under mandated access regulation, investment incentives are
reduced since the agreement to an ex-ante contract is now presented with a forgone
option in the future. As in the case of unilateral investment, coverage decreases with
lower access prices. On the contrary, under a voluntary access regime, coverage is increased at the cost of a less competitive downstream market.
Implications In summary, the literature indicates that the cooperative undertakings
face a similar trade-off between static and dynamic efficiency with regard to access as
in the case of unilateral investment. The introduction of a temporal dimension, how-
36
2.2 Literature survey
TABLE 2.2: Summary - Co-investment.
Effect
Description & References
Static efficiency
Cost reduction
(+)
Collusion
(–)
Downstream Competition
(+/–)
Instrument to lower financing costs, lower up-front
costs per operator but joint production may exhibit diseconomies of scale (Bourreau et al., 2013).
Experimental evidence of tacit price collusion (Krämer
and Vogelsang, 2016).
Multiple firms in downstream market while avoiding duplication and regulation, ex-ante contracts can be used to
dampen competition (Inderst and Peitz, 2012).
Dynamic efficiency
Uncertainty
(+)
Investment incentives
(+/0)
Spreading overall risk facilitates investment (Bourreau
et al., 2013).
Strong incentives for closed cooperatives, weak incentives
for open cooperatives (Rey and Tirole, 2007a). Ex-ante contracts increase coverage and decrease duplication (Inderst
and Peitz, 2012). Coverage only increases if there exists a
demand expansion effect (Bourreau et al., 2013).
ever, may present opportunities to balance this trade-off. The regulator may prescribe
(open) access ex-ante, i.e., before the investment, but not ex-post. The results obtained
by Nitsche and Wiethaus (2011) point towards this direction, but need further assessment and robustness tests in more general settings (Inderst and Peitz, 2012). In particular, this raises the question about adequate parameters to balance this trade-off in
the context of co-investment. For instance, what would be the necessary number of
co-investment participants to secure downstream competition while abstaining from
access regulation. This is related to the more general issue currently debated in European mobile telecommunications market: What is the minimum number of market
participants that is necessary to establish a competitive market? This latter question is
addressed in Chapter 6.
The context of cooperation opens the analysis to a broader behavioral evaluation. The
results obtained by Krämer and Vogelsang (2016) point to several effects that are neglected by theoretical models. This includes especially the analysis of tacit and implicit
collusion during the investment phase and the setting of downstream prices. Identifying the behavioral cause of particular effects may also present new approaches to
37
Chapter 2 A Unified Framework for Open Access Regulation
support regulatory practice. For example, the significant impact of communication itself suggests that improving coordination of investments besides the formation of joint
undertakings may already increase dynamic efficiency. Going further, behavioral approaches may not only be suited to identify these effects, but facilitate the design of
instruments to govern the interactions as will be shown in Chapter 5.
2.2.3 Public-sector participation
In order to provide a brief introduction into the vast literature on the general role of
public-ownership, this sections commences with a reference to an overview of the fundamental theories established within and a summary of several empirical surveys on
the observed effects of public ownership. Thereupon, two more specific issues that have
been related to OA and the deployment of NGANs are reviewed, namely the activities
of municipalities and the formation of PPPs.
Summarizing the fundamental insights of the economic literature on private and state
ownership, Shleifer (1998) concludes that private ownership should be the preferred
option whenever innovation or cost efficiency represent major criteria. The static perspective and the exclusive concern about prices of state ownership deteriorate dynamic
incentives and endanger social welfare in the long-run. Instead, government contracting and regulation are qualified as suitable instruments to ensure public goals that
avoid an excessive role of the state sector. Moreover, Shleifer (1998) refers to the results presented by the public choice theory which highlight the adverse incentives for
government and administration officials under state ownership.
Megginson and Netter (2001) survey the empirical literature on privatization and the
relative performance of privately owned firms compared to state-owned enterprises
(SOEs). Based on studies from numerous countries, which employ different methodological approaches, the authors are able to offer a number of general and robust conclusions. The empirical literature largely confirms that “privately owned firms are more
efficient and more profitable than otherwise comparable state-owned firms" (Meggin-
38
2.2 Literature survey
son and Netter, 2001, p.380). Furthermore, the consensus is that privatization leads
to an improved operating and financial performance of former SOEs. Particularly, output, efficiency, profitability and capital investment spending increases in non-transition
economies, while leverage significantly decreases.
Dewenter and Malatesta (2001) contribute to the empirical evidence that state-owned
firms are significantly less profitable than privately owned firms. The dataset is based
on Fortune magazine’s list of the 500 largest firms worldwide in the timespan from 1975
to 1995. 147 of the 1369 firm-years sample are for government-owned firms, which are
mainly located in Europe. Interestingly, in a survey of privatization procedures no evidence is found that the actual transfer of ownership leads to an increase in profitability.
In fact, profitability is primarily increased prior to privatization, while afterwards no
significant increase can be observed. This points to a structural transformation of the
organization, conducted in advent of privatization, as the actual source of improved
profitability.
Examining the privatization of national telecommunication companies, Bortolotti et al.
(2002) provide specific empirical evidence for the telecommunications sector based on a
global sample of 31 firms in 25 countries. The data set includes full and partial privatizations through public share offerings from 1981 and 1998. In line with the results of the
general empirical literature on privatization, significant improvements of financial and
operating performance are found. Efficiency gains are explained by better incentives
and productivity, while significant cost reductions are also the dominant reason for increased profitability. Again, these improvements cannot solely be attributed to ownership change in most cases, but are also related to accompanying regulatory changes.
In an empirical study of US municipal utilities, Gillett et al. (2006) investigate the determinants that induce these utilities to expand their operations into telecommunications services. The ability to exploit economies of scope stemming from the provision
of internal communications services, and the location in markets with limited existing
competition are found to be significant variables that lead a municipal electric to enter
the telecommunications market. In contrast, rules that aim to impede municipal entry
39
Chapter 2 A Unified Framework for Open Access Regulation
are found to be effective in reducing municipal activity. The finding that rural areas are
less likely to be served by a municipal utility is linked to substantial backhaul costs as
a constraining factor. Due to the ambiguous impact of demographic variables, Gillett
et al. (2006) conclude that the motivation of municipal utilities differs from private organizations. In particular, municipal supply may be motivated by the economic development of financially less attractive areas. Next to their empirical results, the authors
stipulate several arguments in favor of a municipal approach: Utilities are depicted as
early adopters that can accelerate the growth of NGANs and may represent a superior governance mechanism compared to regulation of a private network operator, if
infrastructure-based competition is in fact infeasible.
Ford (2007) presents empirical support for the hypothesis that public investment in
communications networks stimulates entry of private telecommunications firms, as opposed to crowding out private activity. The survey is based on a set of municipalities in
the state of Florida, USA, of which a subset provides electricity and in some cases communications services. Private activity is measured by the number of competitive local
exchange carrierss (CLECs) that are active in the particular market. While the presence of public supply (electricity only) is associated with a lower number of private
communications firms, the provision of communications services by the municipality
increases the activity in the communications sector above the levels of cities that abstain
from self-supply and municipals that provide electricity.
Using a comprehensive dataset that covers the US market from 1998 to 2002, Hauge
et al. (2008) investigate market characteristics that may promote entry by municipal
telecommunications providers relative to the motivation that may induce entry by
CLECs. Based on their empirical analysis, the authors reject the hypothesis that municipal providers base their entry decision solely on expected profits. Interestingly, this
is contrary to what the authors find with respect to CLECs. While CLECs are more
likely to be located in urban areas with higher income and higher revenue possibility, municipal providers tend to serve complementary areas with higher cost structures
and characteristics that predict lower demand. On the other hand, the results indi-
40
2.2 Literature survey
cate a positive, but insignificant relationship between CLEC market participation and
the presence of a municipal provider. Hence, the crowding out hypothesis, that the
presence of a municipal provider impedes entry by a CLEC, cannot be rejected. In conclusion, the authors imply that municipalities do not represent a significant competitive
threat to CLECs and do not impede private market entry.
Troulos and Maglaris (2011), in a qualitative survey of local broadband strategies in
Europe, highlight the ability of municipal initiatives to stimulate demand for NGANs.
The deployment of Metropolitan Area Networks connecting public institutions, implementation of e-Government services, price subsidization, and strengthened civil participation are presented as public instruments to boost network rollout. The role of municipal broadband as a complementary provider of basic infrastructure in rural areas and
as a stimulator of private investments is emphasized. In line with Gillett et al. (2006),
utilities which can exploit economies of scope are shown to be a major driver for the decision to provide municipal communications services. Supplementary, European-wide
subsidies are seen as a public instrument to resolve the fragmentation and heterogeneity created by municipal approaches. Moreover, accompanying state aid regulation
may introduce neutrality as the central principle for supply of access network. On the
other hand, the authors mention the discouraging effect of state-owned network infrastructure on duplication incentives, crowding out private investment at the physical
infrastructure level.
In a meta-analysis, Hodge and Greve (2007) review the effectiveness of PPPs, in particular, long-term infrastructure contracts with a focus on Europe. The analysis is preceded
by a criticism of the extant literature, which highlights the lack of an actual evaluation of PPPs and calls for more empirical analysis which objectively quantifies costs
and gains of such collaborations. Previous studies provide contradictory evidence on
the performance of PPPs, raising doubts whether promised benefits of PPPs can actually be delivered in practice. Moreover, the results suggest that sector-specific variables have a substantial impact on the performance results. This gives rise to concerns
about the accountability in PPPs, due to limited transparency, complexity of the negoti-
41
Chapter 2 A Unified Framework for Open Access Regulation
ated deals and restricted public participation. In addition, administrative and political
decision makers are incentivized by short-term gains, while long term contracts associated with infrastructure projects limit the future flexibility to rectify mistakes or to
adapt to a changing environment. With regard to these issues, the authors highlight
the importance of different organizational structures and the allocation of responsibility. The cross-country analysis indicates that PPPs in the UK and the Netherlands are
implemented as top-down approaches, whereas undertakings in the North European
countries and Germany are organized in a more decentralized way.
Gómez-Barroso and Feijóo (2010) summarize the historical and current debate about
the respective roles of the public and the private sector in the telecommunications domain in their conceptual classification of recent contributions to the debate on privatepublic interplay. The authors state that public activity in the telecommunications industry is supported, although not demanded by the consensus of modern economic
schools, due to the presence of market failures. The hypothesis that markets are in any
case superior with regard to static efficiency is rejected, and instead, a pragmatic logic
(Linder and Rosenau, 2000) is endorsed: PPPs should be considered whenever requirements differ from obviously private or public responsibilities. At the same time, the
authors are cautious about the performance of PPPs, referring to the results by Rosenau (2000) that PPPs display weaknesses in the long-run and by Yescombe (2011) that
PPPs introduce additional complexity.
Developing a theoretical model of incomplete contracts, Hart (2003) draws the conclusion that PPPs should be preferred over individual outsourcing contracts, whenever
it is difficult to specify the quality of the infrastructure, but easy to specify the quality of the services. Since a PPP internalizes the operational expenditures, it engages in
higher productive investment, reducing these costs, but also in excessive unproductive
investment. In the opposite case, when infrastructure quality is easily specified, public
outsourcing yields superior results, since it avoids unproductive investment. In conclusion, Hart emphasizes the role of contracting costs and criticizes the general perception
that the private sector represents a cheaper source of financing than the state.
42
2.2 Literature survey
Based on a conceptual analysis, Given (2010) studies the reinvigorated role of the public
sector and the impact of PPPs in the Australian and New Zealand telecommunications
sectors. He finds that the proposed PPPs in telecommunications are not motivated by
conventional reasons such as private financing and expertise, but rather by the willingness to assert the pursuit of the public interest. Thus, the drastic steps of separation and
nationalization are consequences of the concession that regulation as well as functional
separation have proven to be insufficient instruments in ensuring public goals.
Implications Participation of the public sector in its various forms can boost coverage and speed of network deployment. However, well-known concerns about cost
efficiency, complexity and imperfect information should restrict the scope of public activity. Local initiatives and bottom-up approaches are likely to reduce these mentioned
concerns. Moreover, municipal activity is capable of stimulating demand and extending the scope of viable investment cases due to the consideration of societal spill-over
effects.
Besides the role as a financing facilitator, the public entity is likely to achieve its highest potential at the civil infrastructure level. Since civil infrastructure costs represent
the largest share of total costs (Hoernig et al., 2012), exploiting economies of scope by
providing infrastructure such as ducts can reduce deployment costs significantly.
Mandated OA, as stipulated in the European State Aid Guidelines, can help to keep
distortion of competition through the crowding-out effect at a minimum. In the context
of state-owned access networks, the Australian case will give an indication whether OA
can represent an effective alternative to regulation by ensuring the public interest while
promoting efficient service-based competition on a neutral basis. However, while direct public control may facilitate the implementation of non-discriminatory access, the
literature raises concerns about the adverse incentives of public decision makers and
43
Chapter 2 A Unified Framework for Open Access Regulation
TABLE 2.3: Summary - Public-sector participation.
44
Effect
Description & References
Static efficiency
Economies of scope
(+)
Economies of scale
(–)
Cost efficiency
(–)
Utilities that provide internal communications infrastructure are more likely to expand their services (Gillett et al.,
2006).
Municipal approaches are unlikely to serve rural areas due
to backhaul costs (Gillett et al., 2006).
Private ownership is superior in containing costs (Shleifer,
1998). State-owned firms are found to be significantly less
profitable (Megginson and Netter, 2001; Bortolotti et al.,
2002).
Dynamic efficiency
Stimulation of demand
(+)
Crowding-out effect
(+/–)
Quickness of deployment
(+/0)
Viability of investment
(+)
Hold-up problem
(+)
Innovation
(–)
Regulation
Neutrality
(+)
Accountability
(–)
Public interest
(+/–)
Complexity
(–)
Municipal initiatives can stimulate demand (GómezBarroso and Feijóo, 2010).
Municipal utilities substitutes private market entry (Gillett
et al., 2006). Number of CLECs increases if a municipal provides communications services (Ford, 2007). Municipal providers serve complementary areas relative to
CLECs (Hauge et al., 2008).
Public utilities may serve as early adopters (Troulos and
Maglaris, 2011).
Public investors take into account spill-over effects to adjacent industries (Troulos and Maglaris, 2011).
Anticipation of the long-term socio-economic benefits
(Troulos and Maglaris, 2011).
State ownership neglects dynamic incentives (Shleifer,
1998).
State aid rules as a regulatory instrument to mandate nondiscrimination (Troulos et al., 2010; Troulos and Maglaris,
2011).
Limited transparency and public participation (Hodge
and Greve, 2007).
State ownership as an instrument to ensure public goals
representing an alternative to regulation (Given, 2010).
Adverse incentives of government and administrative officials (Shleifer, 1998).
PPPs “add a substantial layer of extra complexity"
(Yescombe, 2011, p.26).
2.3 Policy guideline
a lack of transparency and accountability in collaborations between state and private
organizations.
2.3 Policy guideline
Up to now the discussion was concerned with the specific effects of a single OA application. Based on these results, this section addresses how the presented applications
relate to each other. The preceding survey of the economic literature has shown that decision makers face multiple trade-offs when deciding for a particular regulatory regime.
Therefore, in the context of NGANs, regulators are required to implicitly or explicitly
weigh conflicting goals. Figure 2.3 illustrates the decision process proposed to assist
regulators in choosing the most appropriate policy instrument, given their primary objective, observed market conditions, and their experience from previously implemented
approaches. Rectangles represent the different regulatory scenarios and diamonds represent the key questions that policy makers must evaluate in order to determine the
next steps and measures in the regulatory decision process. In this vein, the proposed
policy guideline allows for an assessment of the transition between regulatory regimes
when objects are not satisfactorily fulfilled, new problems or technological advancements arise, or the relative order of goals is reevaluated.
In order to make transparent where the identified trade-offs are discussed in the literature, references to representative articles are indicated at the respective decision
branches. Moreover, it is highlighted whether these articles are overviews or whether
they are based on conceptual, theoretical, or empirical analysis. Since the guideline is
derived mainly from the findings within the economic literature, its scope is necessarily
incomplete and policy advice has to take into account complementary legal and technical considerations. Nevertheless, a visual representation and a predefined structure
of the decision process, as laid out in the following, may also facilitate the interdisciplinary discourse between stakeholders.
45
Chapter 2 A Unified Framework for Open Access Regulation
F IGURE 2.3: Recommended policy guideline for OA regulation of NGANs.
Cave (2006)c
Yes, investment by entrants
No
c
Cambini & Jiang (2009)o
Guthrie (2006)
o
Briglauer et al. (2013)
No
Primary
objective?
Lower access price
High retail
prices
Competition
SBC: price regulation
Increased
competition?
No
Primary
concern?
Armstrong (1996, 2002)o
Investment
Yes
Yes, investment by incumbent
e
Bauer & Bohlin (2008)
Effective
FBC?
Klumpp & Su (2010)t
Incentivizes
investment?
Anti-competitive
discrimination
Yes
o
Laffont & Tirole (2001)
Rey & Tirole (2007b)o
t
Vickers (1995)
No regulation
No
Economides (1998)t
t
Mandy (2000)
Yes
Reevaluate
objectives?
No
State aid
Private
investment
ability?
c
Troulos & Maglaris (2011)
c
No
Yes
Enough
investment?
Yes
Goméz-Barrosso & Feijóo (2010)
Yes
Yes
Incentivizes
investment?
No
Multiple
investors?
Nitsche & Wiethaus (2011)t
No
Mandated nondiscrimination
Effective nondiscrimination
Inderst & Peitz (2012)t
Yes
t
No
e
Rey & Tirole (2007a)
Krämer & Vogelsang (2012)
Co-investment
Collusion?
Yes
No
Primary
concern?
Productive & allocative
efficiency
Yes
Enough
investment?
Brito et al. (2012)t
Bourreau et al. (2013)t
No
o
Megginson & Netter (2001)
Bortolotti et al. (2002)e
Shleifer (1998)o
Effective
Non-discrimination
Lafontaine & Slade (2007)o
No
Hardt (1995)c
Iossa & Stroffolini (2012)t
No
Effective
public investor?
Sidak & Crandall (2002)c
c
DeBijl (2005)
Yes
Enough
investment?
Yes
Crandall et al. (2010)e
Ford (2007)e
Hauge et al. (2008)e
Vertical separation
Cadman (2010)c
Riordan (2008)o
Gillett et al. (2006)e
Hodge & Greve (2007)o
State ownership
Private SBC
feasible?
Inderst & Valletti (2009)t
Yes
No
c:
e:
t:
o:
46
Conceptual paper
Empirical paper
Theoretical paper
Overview article, textbook
2.3 Policy guideline
The starting point of the proposed policy guideline is the regulatory status quo, which
may either be constituted by an existing regulatory framework that is potentially to
be replaced or complemented, or by an entirely unregulated industry where new rules
are implemented based on a greenfield approach. Given a status quo, an evaluation
of whether the existing legal and regulatory framework contributes to desireable sector performance is required in the first place. Policy makers will only be willing to
make changes if there is potential room for improvement. The ensuing question of
what can and what needs to be done is the underlying question that is addressed by
the guideline. In any case, it needs to be verified whether (given the status quo) regulatory intervention is (still) justified at all. This is generally deemed not to be the case
if infrastructure or facility-based competition (FBC) is deemed feasible in the relevant
time horizon. This would also imply that the essential facility doctrine does not hold
(Renda, 2010), or in the case of Europe, that the three-criteria-test will fail. In these
cases, deregulation has been suggested to be the optimal choice rather than continued
regulation (Cave, 2006a; Vogelsang, 2013).3
Given the need for regulation, policy makers must now select their primary objective.
This may either be to promote competition (on existing network infrastructure) or to
promote investment into new network infrastructure. Although some scholars argue
that dynamic and static efficiency goals can be reconciled (Klumpp and Su, 2010), a
growing strand of the literature has pointed towards a definite trade-off between these
objectives (Bauer and Bohlin, 2008; Briglauer et al., 2013; Guthrie, 2006). The indication
of slowly expanding NGANs in Europe seems to support the existence of an efficiency
trade-off. Thus, policy makers must inevitably rank either competition or investment as
being more important, while of course, trying to achieve the other as good as possible
under this constraint. In this vein, this branch offers also a historical perspective. On
the one hand, the political and academic debate of the last two decades was centered
around the question of how to introduce and sustain a competitive downstream market
on the basis of existing networks (Armstrong et al., 1996; Laffont and Tirole, 2001; Arm3 Whether infrastructure competition in combination with wholesale competition in fact induces effective
retail competition and thus also benefits consumers under no regulation is investigated in further detail
in Chapter 5.
47
Chapter 2 A Unified Framework for Open Access Regulation
strong, 2002). On the other hand, more recently the need to promote investments in
new NGAN infrastructure is considered to be the main regulatory challenge and, e.g.,
in the context of vectoring (see, e.g., Bundesnetzagentur, 2013a), several regulatory authorities have chosen investment over competition as their primary objective.
If competition is the primary objective, the introduction of service-based competition
(SBC) based on price regulation is proposed as the first regulatory scenario that should be
considered. The policy guideline implies that, in each regulatory scenario, policy makers must continuously evaluate whether the observed or expected market conditions
still warrant to continue the current regulatory approach. In the case of price regulation, such evaluation loops are proposed with respect to investment incentives and
competition. According to the Ladder of Investment principle (Cave and Vogelsang,
2003; Cave, 2006a), appropriate price regulation may incentivize entrants to invest into
new infrastructures themselves. If this occurs, FBC may become feasible after all and
regulation can be lifted. In reverse, even under price regulation, the incumbent may
have a sufficiently large incentive to invest (Klumpp and Su, 2010), which would then
overcome the efficiency trade-off. In this case, there is, of course, no reason to deviate
from price regulation. Similarly, if sufficient competition is achieved, there is no reason to change the functioning regulatory framework. If this is not the case, and the
regulatory objective is primarily to reduce retail prices, then price regulation should be
reevaluated after the access price was lowered. Otherwise, if competition is weak because of anti-competitive (non-price) discrimination, then mandated non-discrimination
should be considered as an alternative regulatory scenario.
If, however, investment is considered the primary objective, the regulator is next faced
with the question whether the private sector is in principle able to provide the necessary investments on its own or whether public-sector involvement is required. Private investment incentives may be fostered by allowing for cooperative approaches
if multiple market participants are able and willing to make partial investments. In
the case of high uncertainty, co-investment may strengthen investment incentives due
to enhanced financing conditions and improved utilization of facilities by infrastruc-
48
2.3 Policy guideline
ture sharing (Bourreau et al., 2013). With regard to competition, co-investment may be,
on the one hand, an instrument to allow for multiple competitors in the retail market
even in the absence of actual FBC or access regulation. On the other hand, regulators should exercise particular caution with regard to collusive behavior in the case of
co-investment as cooperative undertakings provide additional instruments to coordinate behavior among market participants (Krämer and Vogelsang, 2016). In the case of
experienced or highly probable collusion among a closed group of partners, mandated
non-discrimination that allows for OA to the cooperative undertaking may be applied
as an additional criterion. However, regulators need to be aware that these conditions,
again, have a direct effect on the incentives to form such an undertaking in the first
place (Rey and Tirole, 2007a). If co-investment is not a suitable option due to a lack
of multiple investors or due to insufficient investments, mandated non-discrimination as
a stand-alone institution may represent a regulatory alternative to conventional price
regulation.
Mandated non-discrimination comprises regulatory rules that are especially concerned
with providing equal conditions for access seekers relative to subsidiaries of vertically
integrated operators. However, instead of prescribing a uniform (cost-based) access
price, as in the case of price regulation, these rules give operators increased freedom
when setting their wholesale prices. The European Commission’s ex-ante marginsqueeze rule (European Commission, 2013a) can be seen as one possible implementation of this principle. Although this approach is sometimes proposed as a suitable
approach to balance the efficiency trade-off, there is little evidence on the impact of
such rules with regard to the competitiveness of a market and investment incentives.
Hence, further theoretical as well as empirical analysis of the actual implementations
is required. Due to the increased pricing freedom of access providers in this scenario,
regulators need to be particularly cautious about operators’ incentives and abilities to
discriminate against downstream competitors.
If regulators are primarily concerned with issues of anticompetitive discrimination and
less intrusive regulatory rules prove to be ineffective, vertical separation may represent
49
Chapter 2 A Unified Framework for Open Access Regulation
a last resort to eliminate the underlying incentives of anti-competitive discrimination.
In any case, the decision to force a vertical break-up needs to weigh the expected competitive benefits against the potential losses in productive efficiency (Lafontaine and
Slade, 2007) and allocative efficiency (Brito et al., 2012) as well as the affected investment incentives (inter alia, Cadman, 2010; Crandall et al., 2010; De Bijl, 2005). If vertical
separation is still considered the best regulatory option, the question whether the upstream infrastructure should be provided and operated by a private or public entity,
should be guided by an assessment of the investment incentives of private organizations and by the productive efficiency of public investors regarding the specific input
good.
In general, public-sector participation should be considered as a substitute to private
activity only when there is no or insufficient private investment. In the case where
state ownership allows for the exploitation of efficiencies at specific access levels, public activity at these stages may benefit private activity at higher layers (Ford, 2007),
thus representing an exception to the aforementioned general rule. Among the options
for public intervention, state aid represents the less intrusive and thus the primary option, when investment is the regulator’s primary objective and private investments are
not expected. In this case, state aid allows for increased spending and greater public
control, while maintaining market-driven coordination of activities and complementary resources. If state aid is insufficient to provide necessary investment incentives,
state ownership provides a more drastic alternative to increase financing capabilities and
gives public representatives full control of deployment and operations. However, wellknown concerns about low efficiency and financial performance of public investors in
general (Megginson and Netter, 2001), and the success of privatization in accordance
with liberalization in telecommunications in particular (Bortolotti et al., 2002), should
make decision makers skeptical whether the need for higher investments justifies a farreaching conclusion as full public control. When opting for state ownership, efficiency
issues may be mitigated by limiting public activity to the infrastructure level, enabling
service-based competition by private businesses on top.
50
2.3 Policy guideline
In contrast to state ownership, no regulation represents the other extreme on the spectrum of public intervention. Particularly in the case where investment is considered
the primary public objective (and competition is deemed considerably less important),
it may be the only alternative solution to public investment, if the latter is considered
ineffective. By guaranteeing expected rents above the competitive level, investment incentives may be increased compared to competitive scenarios, obviously at the cost of a
loss in static efficiency. According to these implications, the guideline explicitly denotes
whether the regulator is willing to reconsider its goals or whether it is willing to take
such radical measures to stimulate investment.
Of course, at each node, the decision process is not binary in reality, as suggested by Figure 2.3. For example, private investment ability and incentives are likely to differ across
geographical areas depending on household density. Thus, state aid may be made
available for rural areas, but not for urban areas. If there still remain uncovered areas,
public ownership on a municipal level represents a further alternative (Hauge et al.,
2008). Nevertheless, it is often useful to think in dichotomous categories, as depicted
in the guideline, in order to realize the inherent trade-offs. Hence, each regulatory scenario is reached on the basis of some prerequisites and shall be sustained only if this
is warranted by the continuous reevaluation of known issues and expected outcomes.
Otherwise more heavy-handed regulatory scenarios may be considered. Thereby vertical separation and state ownership are seen as the last resort of regulation. In reverse,
if market conditions change, regulators may evaluate the currently implemented regulatory scenario anew, possibly beginning from the starting point of the proposed policy
guideline. Of course, this does not imply that intermediate regulatory scenarios must
actually be implemented. The proposed decision process shall rather be understood as
guiding policy makers to ask the right questions in the right order. Overall, the proposed approach thereby implements a “carrots and sticks" paradigm, equipping the
regulator with more punitive measures if market participants do not comply with the
current set of rules, and offering the ability to lift obligations on the other hand if former
obstacles are removed.
51
Chapter 2 A Unified Framework for Open Access Regulation
The policy guideline also illustrates the numerous reasons that can motivate a mandatory non-discrimination regime, like it is currently pursued by the European Commission (European Commission, 2013a). However, to date there exists only very limited
experience on the expected consequences of this regime. In fact, there is no consensus on what regulatory measures are required to ensure non-discrimination in practice.
While a margin squeeze test could serve as an instrument to counter price discrimination, non-price discrimination is more difficult to monitor (Hardt, 1995). Vertical
separation has been discussed as an effective instrument to counter discriminatory incentives, but at the same time raises severe adverse effects on dynamic incentives as
well as productive and allocative efficiency.
With regard to the role of the public sector at the supply side, the policy guideline
implies that governments should adhere to the principle of subsidiarity. Only when investment is identified as an urgent and absolute primary need that cannot be provided
by the private sector, public ownership on a large scale should be considered as a viable
option. In such a case, vertical separation should restrict public activity to the infrastructure level, allowing for private activity on higher stages of the value chain. The
case of Australia has demonstrated that such a regulatory scenario is a real option. Dissatisfaction with the performance and anti-competitive behavior of integrated network
operators have led the Australian government to take over full control of the access
network infrastructure and to provide connectivity at the last mile as a public service
(Given, 2010). Representing less invasive instruments, state aid, PPP, and municipal
activity can serve as transitory and complementary instruments that should be implemented if they are likely to incentivize further private activity. On the other hand, the
public sector has the general ability to stimulate network deployment by demand-side
measures (Belloc et al., 2012).
52
2.4 Discussion
2.4 Discussion
The investment challenge in the advent of NGANs has introduced a variety of new
business and organizational models as well as regulatory governance mechanisms.
Within the last ten years the role of the public sector in telecommunications has again
taken a turn, increasing its activity through municipal initiatives, PPPs and the deployment of state-owned network infrastructure. In the context of these developments, OA
was suggested by regulators, scholars and industry stakeholders alike as a means to
balance the inherent trade-off between static and dynamic efficiency. However, mutually contradictory interpretations, implementations in different contexts, and a lack of
an explicit definition have prevented a common understanding of the precise meaning
of OA, which in turn may have precluded the rise of such regulation.
By proposing an integrative definition and a structural framework to classify OA application scenarios, this review contributes to a common understanding and a differentiated view on the particular OA concepts. In particular, the survey of the extant
economic literature highlighted that public-sector participation can potentially stimulate network investment and foster private activity. In contrast, concerns about cost
efficiency and accountability should limit state activity to lower access levels where
economies of scope are most significant. In this context, OA regulation can be implemented to minimize competitive distortion on higher access levels. Co-investment may
represent an additional instrument to strengthen investment incentives in cases of high
uncertainty. In this context, OA governing ex-ante access to cooperative approaches
may equip regulators with a new instrument to balance static and dynamic efficiency.
In the case of vertical integration, OA may provide an alternative to price regulation.
In this vein, economies of scope and scale may be exploited while anti-competitive behavior can be contained. At the same time, incentives of non-price discrimination and
the effectiveness of a margin squeeze regulation have to be assessed by the regulator
through empirical analysis. Vertical separation is likely to reduce non-price discrimination, but comes at the costs of structural reorganization and is expected to reduce
incentives for facility-based competition.
53
Chapter 2 A Unified Framework for Open Access Regulation
At last, possible extensions and limitations of the herein proposed conceptual framework and policy guideline are pointed out. First, it must be noted that the presented literature survey and decision framework are necessarily incomplete as they are focused
on regulatory instruments that fall under the proposed definition of OA. For example,
the current outline does not include geographically segmented regulation or regulatory
holidays, which, similar to state aid, co-investment and mandated non-discrimination,
represent additional deviations from traditional access regulation that could foster investment in NGANs. However, it can be argued that these policy instruments are rather
complementary to the considered regulatory scenarios. Hence, it should be possible
to include them in the proposed policy guideline. Yet, research on the interaction between these instruments and the different OA scenarios is scant, particular with respect
to geographic segmented regulation. Thus, future research is required before such an
extension can be proposed comprehensively. Moreover, NRAs need to complement the
presented analysis, which is based on economic arguments, by considering legal and
technical issues that have not been discussed in this chapter.
Second, the proposed policy guideline highlights that there generally exists a trade-off
between static and dynamic efficiency. However, it is worth mentioning that this tradeoff is not necessarily linear. For instance, empirical findings by Aghion et al. (2005)
suggest that the relationship between competition and innovation is characterized by
an inverted-U shape, i.e., innovation is the greatest for intermediate levels of competition. On the other hand, the general validity of this relationship is questioned by other
studies (see, e.g., Sacco and Schmutzler, 2011).
Third, for the sake of clarity, the previous discussion was based on only two of the three
identified dimensions of the OA framework. Evidently, each OA scenario can be subdivided again with respect to the level at which access is provided. Generally, access to
lower network layers (representing high rungs of the ladder of investment concept), allow for a higher degree of quality differentiation and innovation by entrants, since this
enables them to exert more physical control over their resources. Consequently, the access level also has an immediate impact on investment incentives and the scope of non-
54
2.4 Discussion
price discrimination, for example, which may then influence the evaluation at different
nodes of the proposed policy guideline. Currently a set of different technologies and
architectures are implemented in order to realize NGANs, establishing a heterogeneous
landscape at the physical access level. However, at the same time, NGANs implement
IP as the uniform interface at the network layer. This enables widespread availability
of bitstream access, which provides logical unbundling of data flows on the identical physical connection. Since standardization of access products across technologies
may play a large role in platform competition comprising telephony networks, cable
networks and mobile solutions, bitstream access could significantly decrease transaction and integration costs. Furthermore, given the tremendous increase in transmission
capacity that can be achieved by NGANs, there is reason to believe that bitstream access provides a sufficient access level for vital competition and quality differentiation.
Finally, bitstream access as an infrastructure technology-transparent interface may provide a means to specify a harmonized access product across EU member states, which
could represent as a key regulatory instrument to realize an integrated European digital
single market as suggested in the initial proposal for the Connected Continent legislative package by the European Commission (2013e).
Fourth, the discussion of the proposed OA framework was exemplified on the basis
that the monopolistic bottleneck, to which access shall be warranted, is constituted by
the network layer. However, considering the entire ICT value chain, additional bottlenecks may exist further downstream. These may then not be under the control of the
incumbents, but rather be controlled by so-called over-the-top content providers. In
particular, such bottlenecks can be constituted by platforms (e.g., app stores or social
networks). In this context, it is noted that the proposed framework may also be applied
to characterize (open) access relationships at the application layer, using the same basic
dimensions (ownership, vertical structure, access level), but, of course, with different
characteristics of each dimension. For instance, in the case of platforms, the access level
may be differentiated according to features of the application programming interface
(API) that is provided. Nevertheless, mandating access at higher levels of the telecommunications value chain is obviously only warranted in case there is a significant mar-
55
Chapter 2 A Unified Framework for Open Access Regulation
ket failure, e.g., due to inefficient pricing and a lack of replicability. On the one hand,
two-sided platforms have an inherent incentive to price efficiently (cf. Rochet and Tirole, 2006) which implies that market failures are unlikely to occur. On the other hand,
network effects may constitute substantial entry barriers that limit replicability. Thus,
the case for mandating access at the application layer must be closely scrutinized.
56
Chapter 3
Theoretical Foundations and Experimental
Methodology
T
HIS chapter elaborates on the methodological foundations for theoretical and experimental investigations carried out in this thesis. First, Section 3.1 touches on
basic questions in the field of philosophy of science from an applied perspective and
discusses the virtue of theory and theoretical analysis in the field of IS as well as the
interplay between theoretical and empirical research in the context of market design.
Specifically, it is argued that a microeconomically founded research approach is well
suited to guide scientific inquiries in the field of IS in general, and to answer the presented research questions in this thesis in particular. Therefore, an idealized research
process cycle is proposed, which demonstrates how IS and IO research may complement each other in the task to guide the theory-informed design of markets and regulatory institutions. Building on the presented research ideal, Section 3.2 substantiates
how theoretical models and economic laboratory experiments can be employed to serve
as a testbed for the validity and robustness of theoretical predictions as well as new design proposals. Finally, Section 3.3 explores a particular methodological question in the
context of laboratory experiments and examines the various time frameworks that have
been employed in experimental settings. Thus, the section compares decision making
under continuous and discrete time, and in particular considers oligopoly competition
57
Chapter 3 Theoretical Foundations and Experimental Methodology
under continuous time, which allows for real-time, asynchronous strategic interaction.
In order to assess the methodological implications of oligopoly competition in continuous rather than discrete time, a laboratory experiment studies the effects on tacit collusion under both time frameworks, taking into account the mode of competition and
the number of competitors.
3.1 Microeconomically founded research in Information
Systems
Arguably, the main purpose of IS research, like most other research disciplines, should
be the development of robust theories, which can then inform society about the likely
answers to research questions. Notable, although not unique, about IS research is that
the research questions that are pursued are not only concerned with the understanding,
explanation and possibly prediction of real world phenomena, but also with how the
institutions (North, 1991; Roth, 2002) that govern these phenomena in order to achieve
a certain goal (cf. Gregor, 2006) can be shaped. In this regard, IS research takes a theoryguided engineering perspective.
In this vein, design in the IS domain does not only refer to the creation and evaluation of
software artifacts, but encompasses a much broader scope. In particular the economic
context including markets and competitive interactions that evolve around ICT systems
represent a major determinant for the success of information systems (Gimpel et al.,
2008). Taking into account this dimension is vital to reap and maximize the benefits
from advancements of communications and information technology. More specifically,
with regard to the domain of ICT markets, it is of particular interest why an observed
(e.g., technology induced) market behavior occurs, which market outcomes are likely
under a given scenario, but also how markets and the respective regulatory framework
should be designed in order to achieve a desirable outcome.
This section is based on joint work with Jan Krämer (Krämer and Schnurr, 2016).
58
3.1 Microeconomically founded research in Information Systems
In this section, an idealized microeconomically founded IS research process cycle, depicted in Figure 3.1, is presented, which suggests how analytical and empirical research
approaches may complement each other as building blocks to advance robust theory
development. In particular, it is emphasized that fruitful IS theories can be built upon
formal, analytic models. Such models are in turn founded upon both, stylized facts
that are derived from empirical regularities observed in reality, as well as the existing
body of knowledge stemming from robust theories. Here, reality refers to the object
and processes of investigation that research intents to describe or understand. Scientific inquiries are either concerned with realizations of the past or with potential future
states. Researchers perceive reality through empirical observation and data gathering,
which is naturally constrained and imperfect. Models, which in themselves are the
foundation of theory, can then be used to explain, predict and design instances of the
real world. Finally, models, and thus also theory, are evaluated and refined with respect
to their ability to inform society about past or future real world phenomena. This can be
achieved in field or laboratory studies either by validating or falsifying theory-guided
hypotheses, comparing a theory’s predictions with actual future outcomes or by evaluating the success of theory-informed design proposals and engineering approaches in
actual applications.
The herein described research process ideal is more specific than (but not contradictory
to) more general IS research approaches (cf. Frank, 2006), such as design science (cf.,
e.g., Hevner et al., 2004). Nevertheless, it is argued that theories developed under this
framework are suitable to pursue all four fundamental goals of IS research, namely
analysis, explanation, prediction, and prescription/design (cf. Gregor, 2006). It is not
the intention, however, to evaluate or judge different IS research approaches, but rather
to motivate why the proposed microeconomically founded research framework can
serve as one of several appropriate means to rigorously develop relevant IS theories,
in general, and to provide a theoretical foundation for the methodological approaches
employed in the subsequent studies, in particular.
59
Chapter 3 Theoretical Foundations and Experimental Methodology
F IGURE 3.1: Idealized microeconomically founded IS research process cycle.
Deduction &
Induction
Model
Body of
Knowledge
Theory
Hypotheses
Design Proposals
Stylized Facts
Empirical
regularities
Future
scenarios
Empirical
Analyses
Field
Studies
Laboratory
Experiments
Prediction
Future
Data Analytics
Observation
Data Gathering
Reality
Past
Design /
Engineering
Validation /
Falsification
Subsequently, the building blocks of microeconomically founded theory development
are described in further detail: Subsection 3.1.1 elaborates on the conception of theory,
Subsection 3.1.2 reflects on the role of (analytic) models and Section 3.1.3 examines
the role of empirical research methods. Finally, Subsection 3.1.4 discusses concerns of
whether there is in fact too strong of an emphasis on theory in scientific practice (cf.
Hambrick, 2007).
3.1.1 Theory as a set of models
In general, theory has been characterized as the “basic aim of science”(Kerlinger, 1986,
p.8) and is often referred to as “the answer to queries of why” (Kaplan and Merton cited
by Sutton and Staw, 1995, p.378). According to Weick (2005) a theory may be measured
in its success to “explain, predict, and delight” (p.396).
The precise understanding of theory here is based on the premise that the main task
of theory is the integration of findings of individual studies into a modular, but coherent body of knowledge that connects research agendas based on a shared terminology
60
3.1 Microeconomically founded research in Information Systems
and which provides a microfoundation. Revision and extension of theory is achieved
in iterative steps through new or modified models that may either re-investigate central assumptions, thus deepening theory’s microfoundation, or create meta-models by
further abstraction based on the existing body of knowledge. By this means, a microfounded theory serves as an anchor (Dasgupta, 2002) and provides building blocks for
new research projects and further theory-building.
According to this view, robust theories are the result of deduction and induction from
a host of formal models. Therefore, theory can be viewed as a classified set or series
of models (Morgan and Knuuttila, 2012). In philosophy of science this integral role of
models as a part of the structure of theory has been supported by the Semantic View and
has been further emphasized by the Pragmatic View (Winther, 2015). Consequently, a
clear distinction between theory and its models is difficult in general, and even more so
if the analysis of theoretical models is deemed as the central part of scientific activity.
At the extreme, a single model may already be the foundation of a theory, although
probably not a very robust one. In this regard, the understanding of a robust theory in
the social sciences may differ from the understanding of a robust theory in the natural
sciences, because theory in the social sciences can be very context dependent, as subjectivity of decision makers, i.e., their beliefs, information, and view of the world substantially shape their choices and actions (Hausman, 2013). For example, Dasgupta (2002)
noted that “the physicist, Steven Weinberg, once remarked that when you have ‘seen’
one electron, you have seen them all. [...] When you have observed one transaction,
you have not observed them all. More tellingly, when you have met one human being,
you have by no means met them all” (p.63). This is why a robust theory in the social
sciences should regularly be built upon a set of models, each of which takes a different
perspective on a particular issue and explores a slightly different set of assumptions,
such that the boundaries of the theory become transparent.
61
Chapter 3 Theoretical Foundations and Experimental Methodology
3.1.2 Models as mediators between theory and reality
This understanding of theory shifts the attention to the development of suitable models. Models as idealizations (Morgan and Knuuttila, 2012) serve as representations of reality that are obtained by simplification, abstraction (see, e.g., the work of Cartwright,
2005; Hausman, 1990) and/or isolation (Mäki, 1992, 2012). But they may also be created
as pure constructions, i.e., exaggerated caricatures (Gibbard and Varian, 1978), fictional
constructs (Sugden, 2000) or heuristic devices that “mimic [...] some stylized features of
the real system” (Morgan and Knuuttila, 2012, p.64). Gilboa et al. (2014) suggested that
economic models serve as analogies that allow for case-based reasoning and contribute
to the body of knowledge through inductive inference rather than through deductive,
rule-based reasoning. The use of formal, analytic models in this context can be advocated, because such models allow to make the assumptions transparent that may lead
to a proposition and possibly a normative statement upon which a robust theory, and
ultimately a robust explanation or prediction can be built. Note that mathematical formalization is a sufficient, but not a necessary prerequisite to develop a formal model,
because it allows to precisely formulate its subject domain, making it an “exact science”
(Griesemer, 2013, p.299). Moreover, Dasgupta (2002) argued that in building a theory
“prior intuition is often of little help. That is why mathematical modeling has proved
to be indispensible” (p.70f). The analytic approach provides researchers with a toolbox
to deal with especially hard and complex problems. By the means of logical verification, propositions can be shown to be internally true with regard to the underlying
assumption.
In general, the goal of a model is to “capture only those core causal factors, capacities
or the essentials of a causal mechanism that bring about a certain target phenomenon”
(Morgan and Knuuttila, 2012, p.53). Such an abstraction is the prerequisite for conducting a deductive analysis within a particular scenario of interest. What is considered
to be particularly important in order to develop relevant models is that a model’s microfoundation should contain elements of both theory and reality. On the one hand,
a model’s assumptions should reflect stylized empirical facts that are well grounded in
62
3.1 Microeconomically founded research in Information Systems
observed empirical regularities or relevant future scenarios. Such empirical facts can be
derived directly from gathered data (most likely with measurement error), may already
be the result of extended data analysis, e.g., in the form of detected patterns or correlations, or may be identified by means of a literature review (Houy et al., 2015). However,
stylized empirical facts need not (yet) be supported by any theory. Thus, it is also possible to incorporate insights of theory-free empirical analysis (particularly (big) data
analytics or machine learning) into formal models, which may then lead to a theory
that can explain the empirical regularities.1 On the other hand, a model’s assumptions
may also be derived from the existing body of knowledge, i.e., from theory. This exemplifies the dual view on the relationship between models and theory: Although models
are used to advance theory, theory is also used to produce and inform models.
A main line of attack against analytic models is to argue that they are not realistic and
thus, model-driven theory is useless, because there is nothing to learn about reality.
This criticism is amplified in the field of social sciences, where models are context dependent, as argued above. This naive understanding, however, falls short. First, as
mentioned above, “good” models should be grounded in stylized empirical facts. Second, there is an inherent trade-off between accuracy and generality, achieved trough
simplicity (Gilboa et al., 2014). Scholars experienced in the domain of modelling generally agree on the fact, that too much complexity in fact impedes the explanatory power
and the interpretability of models. For example, Schwab et al. (2011) stated that in order
“to formulate useful generalizations, researchers need to focus on the most fundamental, pervasive, and inertial causal relations. To guide human action, researchers need to
develop parsimonious, and simple models that humans understand” (p.1115). In the
words of Lucas (1980) “a ‘good’ model [...] will not be exactly more ‘real’ than a poor
one, but will provide better immitations” (p.697). In this context, the statistician George
1 In
this context, it is worth mentioning that although data analytics may be able to predict what will
happen in a specific context, similar to a theory, it is still theory-free, because it is generally not able to
explain why it happens. Without theory, however, it must remain unknown whether these predictions
can be generalized and and to what extent they are robust to other application scenarios. Therefore,
data analytics differs from the traditional paradigm of empirical analysis, which centers around the
falsification or validation of hypotheses, which again requires a theory (although not necessarily in
the same sense as proposed here - see, e.g., Diesing (2008) for a more elaborate discussion of the relationship between empirical and formal theory) from which these hypotheses are derived in the first
place.
63
Chapter 3 Theoretical Foundations and Experimental Methodology
Box coined the famous phrase that “all models are wrong, but some are useful” (Box,
1979, p.2), clarifying that a model must inherently be unrealistic in a dogmatic sense
(see Mäki, 2012, for a discussion), but that models in fact enable one to understand real
phenomena by abstracting from the complexity of reality. To exemplify this, Robinson
(1962) argued that “a model which took account of all the variegation of reality would
be of no more use than a map at the scale of one to one” (p.33). Of course, an interesting model must also exceed a pure tautology, i.e., the results that can be deduced
from its assumptions are usually not a priori clear, but may represent surprising results
(Koopmans, 1957; Morgan and Knuuttila, 2012). This requirement can be paraphrased
by a quote that is supposedly due to Einstein: “Everything should be made as simple
as possible, but not simpler” (cf. O’Toole, 2011).
Furthermore, it is emphasized that over and beyond the explanatory function of formal
models, the modelling process itself may prove to exhibit value for understanding a
particular scenario. Moreover, a model is an instrument to express an individuals’ perception of a problem and may therefore serve as a communication device. Gibbard and
Varian (1978) stated that “perhaps, it is initially unclear what is to be explained, and a
model provides a means of formulation” (p.669).
3.1.3 Empirical analyses as the means to evaluate theory
According to this theory-centric research view, empirical analysis serves two core functions: i) As described above, empirical analysis is a means to derive stylized facts in
order to motivate model assumptions, or likewise, to evaluate the plausibility of proposed assumptions. ii) As will be described next, empirical analysis is also a means to
evaluate the quality of a theory as a whole. In the context of IS research, three main
ways in which evaluation of theory can be done are conceived.
First, empirical analysis, foremost field and laboratory studies, can be employed in order to falsify (in the spirit of Lakatos and Popper (Hausman, 2013; Backhouse, 2012)),
and more ambitiously to validate, theoretically derived hypotheses. Whereas field
64
3.1 Microeconomically founded research in Information Systems
studies have the advantage of high external validity, they can be generally challenged
on the premise that it is difficult to establish causal effects due to problems of (unobserved) confounding variables and endogeneity. At a fundamental level, this gives
rise to doubts whether empirical observations are able to falsify (a fortiori validate)
theory at all. These concerns are magnified due to the context-specific nature of field
studies and a lack of control over the environment that encompasses investigations.
Laboratory experiments may be able to mitigate some of these concerns through systematic variation of treatment conditions, randomization of subjects and augmented
control of the researcher. Based on a high internal validity, although at the cost of lack
of external validity, isolation of causal relationships is facilitated and falsification of
theoretical propositions is more easily justifiable (Guala, 2005). Furthermore, laboratory experiments facilitate the process of de-idealization (Morgan and Knuuttila, 2012),
i.e., the generalization of the model context beyond its well-defined assumptions by
successively relaxing the assumptions until the theory’s established hypotheses begin
to break down. Ultimately, however, laboratory and field studies are complementary
means to a similar end.
Second, empirical analysis can evaluate the accuracy of theory-driven predictions over
time. Although hypotheses may also be regarded as model predictions, the focus here
lies less on falsification of suggested causal relationships, but more on the correct qualitative assessment of the impact of future scenarios. With regard to its ability to predict
future states of reality (in the sense of Friedman, 1953), a microfounded theory draws
from its ability to explain observations at the macro level, based on an understanding
of the underlying mechanisms and the necessary conditions. By this means, theorydriven predictions are likely to be more robust to changes of real systems as underlying causes can be identified and theory can be modified accordingly (Dasgupta, 2002).
Moreover, formal analysis allows for experimentation and evaluation of counterfactuals. Two remarks should be made in this context: First, it must be noted that there exists
an inherent trade-off between a theory’s simplicity and its predictive accuracy. While
a simple model or theory may apply more generally and is able to make more robust
qualitative predictions, it will also almost certainly be too simple to make accurate quan-
65
Chapter 3 Theoretical Foundations and Experimental Methodology
titative predictions. In turn, the reverse holds true for complex models. This is akin to
what is known as the bias-variance-trade-off in statistics (cf. Hastie et al., 2009). Second, even if a theory’s prediction may be accurate, this does not “prove” in a deductive
sense that it is valid. One may only apply what is known as abductive inference here,
that is one can infer that a theory was sufficient to predict the phenomenon of interest,
but not that it was necessary, i.e., the only possible theory to be sufficient.
Third, and particularly relevant for this thesis, empirical studies can serve as a testbed
for theory-driven design proposals. In this context, laboratory experiments can be seen
as an intermediate economic engineering step, similar to a wind tunnel in traditional
engineering, where the design proposals (e.g., a proposed market design or regulatory
institution) can be evaluated under idealized conditions that mirror those assumptions
under which the theory was developed. If the proposed design performs well (relative
to the intended goal) in the laboratory then it should be taken to the field for further
evaluation. If, however, the proposed design already fails to perform in the laboratory,
then there is little reason to believe that it would perform well in the field (Plott, 1987).
Consequently, the design, and most probably also the underlying theory, would need
revision already at this stage.
3.1.4 Discussion
Several scholars in the fields of management (Locke, 2007; Hambrick, 2007) and IS (Avison and Malaurent, 2014), among others, have criticized excessive adherence to theory
and argue that a scientific contribution can also be made without the need for theory.
Whereas this raises important issues with regard to the research process and scientific
institutions that are necessary to develop theory, these views still can be reconciled with
the belief that the development of robust theories is at the core of scientific endeavour.
Above and beyond, the arguments set out above suggest that these models and theories
should be both, i) well grounded in stylized empirical facts that are the result of inductive research efforts, as well as ii) evaluated and refined through empirical analyses
66
3.2 Theoretical modeling and experimental evaluation
based on field studies and laboratory experiments. To this end, a microeconomically
founded IS research process ideal that is deemed suitable to develop theories that are
rigor and relevant has been motivated and discussed. In this spirit, the long term goal
of microeconomically founded IS research is the development of robust and stable theories that have been established and refined through several repetitions of the depicted
research process cycle.
3.2 Theoretical modeling and experimental evaluation
Technological innovations, new business models, and adjusted policy objectives have
significant effects on firms’ strategic incentives, on the market structure, and thus on the
competitiveness of markets as outlined in the preceding chapters. In consequence, established grounds for ex ante regulation and ex post competition policy may no longer
be justified. Thus, sector-specific rules such as in the telecommunications industry need
to be repeatedly evaluated and adjusted accordingly (see the current review of the European framework by the European Commission, 2016c). Similarly, competition policy,
which closely interacts with the regulatory regime specifically in the case of merger control and antitrust, needs to adjust to a changing environment. Although competition
authorities decide on a case-by-case basis, investigations and verdicts follow general
policy objectives, assumptions, and procedures, which are explicitly codified in guidelines or implicitly given by the history of previous decisions (for evidence that decisions
are not exclusively determined by case-specific facts and considerations see, e.g., McGowan and Cini, 1999; Duso et al., 2007). Therefore, both regulatory and competition
authorities require systematic and robust evidence on the effects that result from these
changes under specific market characteristics and alternative institutions.
Authorities, assigned with the task to identify and design new regulatory mechanisms
in the light of these environmental changes, face significant challenges. On the one
hand, they have to assess the impact of changing external conditions, while also taking
into account the probability of further change. Thereby, the effects of external factors
67
Chapter 3 Theoretical Foundations and Experimental Methodology
may not always be easily inferred from the observed outcomes in practice, because
these effects likely interact with the current policy framework. On the other hand, regulatory mechanisms—designated to address the new circumstances—may be untested,
as they have just become available due to technological innovation (see for example the
Layer 2 bitstream access product for cable networks standardized by the Bundesnetzagentur, 2013b) or fundamental changes in the market structure. Even if institutions
have already been implemented in other contexts, e.g., in another geographical region,
it is often unclear whether implications transfer to the considered scenario of interest. In summary, high uncertainty together with high societal costs in case of a failed
implementation call for the theoretical investigation and the controlled assessment of
such policy proposals and regulatory institutions prior to the implementation in the
field. In this vein and in the spirit of Section 3.1, analytic modeling and experimental
evaluation are proposed as the primary research methods in this thesis to examine the
performance of alternative regulatory institutions under varying market structures and
different market characteristics.
Based on the proposed research process ideal, this section first expands on why analytic
(game-theoretic) models are well suited to address the research questions in this thesis.
Therefore, on the basis of Morgan and Morrison (1999) and Morgan (2002, 2005), the following four functions of models in the context of theoretical and experimental analyses
are discussed: 1) models as representations, 2) experimentation on models, 3) theory testing,
and 4) experiments with models. Subsequently, the use and advancement of models in
this thesis is outlined. Finally, the shortcomings of a purely theoretical model-based
approach are highlighted and it is discussed how those issues may be resolved by complementary experimental analyses.
1) Models as representations and a collection of stylized facts
Whereas new models
may also be deduced based on theoretical considerations exclusively (Friedman, 1953;
Wong, 1973), this thesis is concerned with the design and evaluation of institutions
in an applied context. Therefore, models as mediators (Morgan and Morrison, 1999,
Chapter 2) combine theory, i.e., the existing body of knowledge, with empirical regu-
68
3.2 Theoretical modeling and experimental evaluation
larities (stylized facts), i.e., observations of reality. In consequence, a model implicitly
represents the decisions about what is deemed relevant about a particular delineated
aspect of reality from a theoretical point of view. As mentioned in the previous Section,
this gives rise to the trade-off between representational accuracy and generality, but in
practice also tractability (Lawson, 2008). Whether the model adequately captures and
thus represents the relevant characteristics of reality in its assumptions critically determines the power of back inference, i.e. whether results and implications from a model
carry over to reality (Morgan, 2002).
2) Experimentation on models Morgan (2005) characterizes models as autonomous
instruments of investigation that allow for experimentation. In this vein, “modeling
work is creative and exploratory” (Morgan, 2002, p.50) as a model based on a set of
axioms and assumptions can be examined with respect to its attributes and interior
mechanics, the outcomes and solutions that can be derived, and its generality and limits. Thus, the manipulation of economic models—representing a set of stylized facts of
the status quo—allows to assess modifications of established institutions theoretically
and to compare hypothetical counterfactuals. The creation of an artificial world and
its analysis (Morgan, 2002) constitutes a basis to address research questions even if no
empirical basis is available. However, the (external) validity of the obtained results
and inference from the model hinges critically on the realism of the assumptions that
are most relevant to the investigated scenario. As argued above, the representativeness
of a model can thus have direct ramifications for the epistemic power of its deduced
results, emphasizing the importance of aligning a model’s assumptions with empirical
stylized facts.
3) Empirical theory testing on a model basis
Given the research process ideal out-
lined in Section 3.1, the validity of a model’s assumptions and the robustness of its
results are subject to empirical evaluation. Economic laboratory experiments allow for
a high degree of control regarding the variation of treatment variables and the randomization of influencing (confounding) factors, which yields a high internal validity
69
Chapter 3 Theoretical Foundations and Experimental Methodology
(Guala, 2005). Hence, experiments are well suited to provide a first test on whether
the model’s hypotheses hold under the stated assumptions if they are transferred from
an artificial world to a domesticated version of reality (Harré, 2003). Furthermore, the
experimental design allows to test the robustness of insights and to isolate the causal
factors that yield a particular result by incremental and factorial treatment comparisons. Moreover, experimental studies are able to quantify effect sizes and trade-offs,
which is often infeasible based on purely theoretical work with abstract models. Obviously, the specification of the experimental design and the transformation of an abstract
model into an operational format require parameterization, i.e., a de-idealization of the
underlying generalizing assumptions (Morgan and Knuuttila, 2012). Thus, the design
of an experiment offers a considerable degree of freedom over and beyond the underlying model specification. Naturally, these specific design decisions limit the theory test
to a narrower application scenario than that for the abstract model, which in turn may
question the robustness of the findings of an individual study. In combination with
concerns regarding the validity of statistical assessments, this has raised momentum
for replication studies and meta-studies in several scientific disciplines (see Chapter 8
for a concluding discussion). In this vein, Chapter 6 conducts a meta-study of number effects on the competitiveness in oligopolistic markets which integrates individual
studies that are based on different model assumptions. On the contrary, degrees of freedom with respect to the parametrization of an experimental design can be exploited to
control for behavioral effects, which are considered out of scope for the particular test
of theory. For example, all experimental studies in this thesis parameterize the underlying competition model in a way that participants face identical graphical interfaces
across treatments in order to avoid behavioral framing effects.
4) Experiments with models Analytic models may be employed as experimental instruments even when the model itself is not the object of investigation (Morgan, 2002).
Thereby, some outcome or process of an experiment is realized by the instantiation of a
theoretical model instead of the controlled interaction between experimental subjects.
For instance, in an experimental market subjects may only represent sellers, whereas
70
3.2 Theoretical modeling and experimental evaluation
demand is computed based on a theoretical model of a representative consumer that
reacts to the sellers’ offers. Here, behavior and outcome is controlled by the model’s assumptions instead of experimental variation (Morgan, 2002). Hence, experiments with
models may be particularly useful if the investigated scenario is characterized by high
complexity, and thus experimental control would be difficult to ensure. In this vein, all
experimental studies in this thesis employ stylized competition models as experimental
instruments and thus explicitly abstract from details of real transaction processes.
Use and advancement of models in this thesis
Based on these diverse functions, this thesis draws on oligopoly theory and employs established models of strategic price and quantity competition to investigate the research
questions of interest. Theoretical analyses in Chapter 4 and Chapter 5 examine access
relationships based on a model of vertically related upstream and downstream markets, building on the recent literature on wholesale competition (Bourreau et al., 2011,
2015; Höffler and Schmidt, 2008). In these models, the retail demand structure is based
on the representative consumer suggested by Shubik and Levitan (1980), whereas homogenous Bertrand competition is employed at the upstream level. Exploring these
models analytically (in the sense of Morgan, 2005), this thesis adds to the theoretical
body of knowledge by analyzing Open Access regulation regimes under infrastructure
and wholesale competition. The models’ theoretical equilibrium predictions with respect to wholesale competition and open access regulation are then tested experimentally (Chapter 5). Experimental analyses of oligopoly competition in Section 3.3 and
Chapter 6 employ models as experimental instruments and exploit the duality of the
theoretical demand model proposed by Singh and Vives (1984). To this end, the experimental studies build on the model’s generality, its well-known theoretical predictions
and its previous application in numerous theoretic analyses (inter alia López and Naylor, 2004; Garella and Petrakis, 2008) as well as in an experimental meta-study (Suetens
and Potters, 2007).
71
Chapter 3 Theoretical Foundations and Experimental Methodology
Whereas the theory on oligopolistic competition in one-sided markets is fairly mature
due to its long history and central role in economic literature, theory-building in the
context of digital media and services markets is still in a relatively early stage, although
advancing quickly in the recent decade. In particular, competition between online content platforms is likely to be multi-sided (Rochet and Tirole, 2006; Armstrong, 2006;
Hagiu and Wright, 2015) due to the presence of indirect network effects (cf. Parker and
Van Alstyne, 2005) and users’ ability to multi-home, i.e., to use services of different
providers at the same time (cf. Choi and Kim, 2010). These characteristics are especially
relevant for the competition between online content providers that monetize users’ eyeballs through advertising (Evans, 2008). Although there is a rich theoretical literature
on advertising competition in offline markets (see, among others, Anderson and Coate,
2005), there is currently no model framework for online advertising competition that
easily relates online advertising prices and publishers’ ability to track and target consumers. Thus, Chapter 7 offers a theoretical microfoundation for this scenario based on
stylized empirical facts and thereby proposes a new model building block to the growing theoretical literature on the competitive effects of data collection and data analysis
in online markets (see, e.g., Athey and Gans, 2010; Asdemir et al., 2012; Athey et al.,
2014; Montes et al., 2015).
Complementary use of models and laboratory experiments in this thesis
As outlined in Section 3.1, theoretical model analysis and empirical evaluation naturally complement each other as research methodologies. Conversely, several issues
arise if the analysis is exclusively restricted to a single approach. In this vein, the main
shortcomings of game-theoretic solution concepts, which are regularly utilized in the
theoretical analysis of strategic settings, are highlighted in the following. First, behavioral aspects that are neglected by conventional (one-shot) equilibrium solution concepts may considerably affect market outcomes. Specifically, tacit collusion, which has
been found to significantly affect firm behavior in markets with few competitors (Parker
and Röller, 1997; Engel, 2007; Potters and Suetens, 2013) and is deemed a critical factor
72
3.2 Theoretical modeling and experimental evaluation
in the oligopolistic telecommunications markets considered here (cf. BEREC, 2015), is
currently only rationalized in an infinitely-repeated game context by mainstream economic theory (see, e.g., Nocke and White, 2007; Normann, 2009). Yet, it is well-known
that coordinated behavior is also frequently experienced in experimental oligopoly settings that resemble a finitely-repeated game context (see, e.g., Huck et al., 2004b; Orzen,
2008). Moreover, theoretic analyses may predict a multiplicity of equilibrium outcomes
for a given scenario and model (see, e.g., Bourreau et al., 2011, for multiple equilibria in the context of wholesale competition). In such cases, it often remains unknown,
which of the identified equilibria is more likely to arise. It is thus infeasible to quantify
or at least indicate the probability to arrive at a particular outcome in practice. The
equilibrium selection problem may be reduced or solved by introducing additional assumptions. In many cases, however, this would only defer the problem to a preceding
stage of the modeling process, because it is still unclear which assumption indeed applies to the context of interest. In fact, it may more generally be even unknown which
solution (equilibrium) concept is most adequate in the respective scenario (cf. Binmore,
1987). Furthermore, game-theoretic equilibrium solution concepts per definitionem neglect out-of-equilibrium behavior as part of a possible outcome, which may considerably impact strategic behavior and outcome in actual market practice (cf. Arthur, 2006).
Although there are theoretical concepts to test the robustness of equilibrium outcomes,
such as trembling-hand equilibria (Selten, 1975) and quantal response equilibria (McKelvey and Palfrey, 1995), these concepts in turn require additional assumptions about
such behavior. In summary and more abstract terms, the listed issues stem from one
of the main advantages of theoretical models as artificial worlds: the explicit and total
control over the assumptions. Either these must be explicitly and completely specified
a priori or the model may abstract from specific issues entirely, but there is never a
“theoretical” approach to let them be determined implicitly.
In contrast, laboratory experiments that provide the “potential for independent action”
(Morgan, 2005, p.325) due to a controlled “degree of freedom on the part of participants” (Morgan, 2005, p.325) can leave assumptions unspecified and let them be determined implicitly in the experiment. On this basis, Harré (2003) argues that exper-
73
Chapter 3 Theoretical Foundations and Experimental Methodology
iments allow for stronger back inference. In turn, this leads Morgan (2005) to argue
that economic experiments exhibit greater epistemic power than theoretical models,
because experiments may yield “results that may be unexplainable within existing theory” (p.317). According to this view, experiments may not only falsify theory or provide
empirical support, but may represent—like theoretical modeling—an “explorative creative activity” (Morgan, 2005, p.318), which yields unexpected observations that enable
the conceptualization of new theories. Moreover, laboratory experiments as a complementary methodology can be exploited to refine theoretical predictions, especially in
the case of multiple equilibria (cf. Abbink and Brandts, 2008). Laboratory experiments
according to this understanding may thus serve two purposes within the idealized research process cycle (cf. Figure 3.1): Empirical validation or falsification of theoretical
hypotheses as well as observation of (isolated) empirical regularities, which inform the
formulation of new or refined theories.
Laboratory experiments are at the core of scientific endeavor, most notably in the physical and life sciences. In comparison, economic laboratory experiments have emerged
relatively late, but their role and adoption have significantly grown over the recent
decades (Falk and Heckman, 2009). Due to high internal validity (Guala, 2005), they
are well suited to identify systemic effects and to isolate and measure their causal determinants through controlled variation of exogenous variables. Firm and oligopoly
experiments, as surveyed by Potters and Suetens (2013), have therefore been established as an important research approach to test theoretical predictions and to collect
observations on empirical regularities in the context of strategic competitive interaction. Moreover, laboratory experiments are utilized to examine the policy implications
of institutions which may have not (yet) been implemented in the field and to benchmark the results against designed counterfactuals (see Normann and Ricciuti, 2009, for
a survey of laboratory experiments that address economic policy issues). In this spirit,
the experimental studies in this thesis make use of the freedom with regard to the experimental design as well as with regard to experimental behavior of participants to address
the research agenda outlined in Chapter 1. In particular with regard to the issue of tacit
collusion, which is notoriously hard to detect in field studies, laboratory experiments
74
3.2 Theoretical modeling and experimental evaluation
can provide insights by analyzing in- and out-of-equilibrium strategies and respective
market outcomes relative to benchmark equilibria predicted by economic theory. In
this vein, economic laboratory experiments may be seen as a cost-efficient testbed for
the evaluation of theoretical hypotheses and the comparison of alternative mechanisms,
which is particularly relevant in the context of regulatory institutions.
Whereas theoretical models may suffer from “problems of realism” (Morgan, 2005,
p.321), experiments may lack external validity, i.e., it remains unclear whether the experimental results apply beyond the laboratory context in the relevant domain of investigation (see Schram, 2005, for a discussion). Despite findings that offer support for the
external validity of economic experiments with student subjects (Ball and Cech, 1996),
there remain concerns particularly with respect to laboratory experiments that study
strategies and behavior at the firm level. Most commonly, critics suppose that laboratory experiments are not able to capture the complexity of “real markets”. Instead, empirical evaluation should be carried out immediately in the field. However, as pointed
out before, there often may exist no field implementation with regard to the issues addressed by this thesis. Moreover, field studies face several challenges themselves with
regard to the generalizability of findings and endogeneity concerns as further discussed
in Section 6.1. In fact, the lack of control and randomization in many conventional field
studies have introduced the experimental design as a desired benchmark and led to
the emergence of empirical studies that exploit quasi-experiments (Angrist and Pischke, 2010). In summary, these issues highlight the complementary nature of empirical
evaluation methods in contrast to the notion of a unique gold standard. In the interest
of robust theory-building it therefore seems valuable to employ heterogenous empirical methods, which each may examine different aspects of a theoretical prediction. In
this vein, methodological limitations together with the implications for future empirical
work for each individual study will be provided in the respective chapters. Finally, concerns regarding the external validity of laboratory experiments are further addressed
by a validation study in Chapter 5, which replicates an experimental treatment with
expert subjects and compares results to the student sample.
75
Chapter 3 Theoretical Foundations and Experimental Methodology
In a further effort to foster external validity and to facilitate decision making in complex
strategic settings, the study on wholesale competition in Chapter 5 employs an experimental framework in continuous time. In contrast to the vast majority of oligopoly
experiments, which have employed experiments in discrete time, a continuous time
framework does not prescribe any exogenous timing order nor the number of actions
that can be undertaken by the subjects. In the spirit of Morgan (2005), this design feature is believed to foster epistemic power because it endogenizes the decision over the
timing of actions and thus extends the degree of freedom for subject behavior. By abstaining from a fixed timing structure, experiments in continuous time are able to test
the underlying theoretical model’s assumptions of a specific timing structure. Moreover, by allowing for continuous decision making and thus the freedom to constantly
observe, revise and adapt decisions, subjects’ task to understand and act in complex
experimental settings is arguably facilitated. Discrete time frameworks inherently restrict the number of actions and often employ only a relatively low number of periods,
which may lead to findings that rather resemble laboratory artifacts than stable longterm behavior (see, e.g., Friedman et al., 2015; Oechssler et al., 2016, for a comparison of
short-term and long-term experiment horizons in a discrete setting). In contrast, continuous time frameworks do not constrain the number of decisions and may thus also
encourage exploratory strategies. This may be of particular importance if theoretical
models predict a multiplicity of equilibria and the experimental design needs to ensure
that all equilibria are in fact discovered by the subjects.
Despite these advantages, there may be concerns that the design of experiments in continuous instead of discrete time may significantly affect subjects’ behavior and thus
influence experimental findings in an unknown direction. Yet, only few studies have
investigated the effect of an experimental real-time setting that runs continuously and
in which players, per definitionem, hold onto their actions until they change them explicitly. Comparative studies have been conducted in the context of prisoner’s dilemma
games (Bigoni et al., 2015; Friedman and Oprea, 2012) as well as in a Hotelling (1929) location model (Kephart and Friedman, 2015; Kephart and Rose, 2015). However, no systematic investigation on the effect of continuous versus discrete time framing on price
76
3.3 Oligopoly competition in continuous time
and quantity competition in oligopoly experiments exists to date. Therefore, the following section scrutinizes whether prices and profits differ in discrete and continuous
time competition, i.e., whether one of the two time frameworks facilitates cooperation
and the ability to tacitly collude relative to the other.
3.3 Oligopoly competition in continuous time
Decision-making is, by its nature, a continuous process. Individuals but also organizations monitor their environment continuously and may act or react according to their
observations at any time. Especially in electronic markets, e.g., online retail or financial
markets, sellers and buyers may react promptly to decisions by other market participants. However, the reaction time by decision makers can also be chosen strategically
(e.g., strategically delayed), or actions may be taken only for a very short period (e.g.,
to send a “signal" to the competitor or to retaliate a competitor’s action). In continuous
time, the reaction time or duration of an action is chosen endogenously by the decision
makers and thus offers a richer set of strategies than if actions can only be taken at fixed
points in time.
However, most economic laboratory experiments have so far employed a discrete time
framework and have thus relied on the assumption that players move simultaneously
or sequentially in a pre-defined and ordered sequence. Consequently, subjects in the
experiment have a given (limited or infinite) amount of time to decide on their actions.
By this means, these experiments abstract from the decision makers’ choice with regard
to the timing of an action and thus implicitly restrict the space of potential strategies.
This study examines the effects when the restrictions imposed by a discrete time framework are removed and decision making is allowed to take place in a continuous time
framework. Thereby, this study makes the following two contributions to the literature: First, it provides an examination of non-cooperative game settings in continuous
time, a time framework that captures more elements of reality than discrete time but
This section is based on joint work with Niklas Horstmann and Jan Krämer (Horstmann et al., 2015).
77
Chapter 3 Theoretical Foundations and Experimental Methodology
for which theoretical predictions are scarce and ambiguous. Second, it is tested empirically whether the degree of tacit collusion differs between experiments in discrete and
continuous time. By doing so, the results may inform future experimental designs as
well as the appraisal of previous findings from discrete time experiments on oligopoly
competition.
Taken to the laboratory, continuous time implies that the length of a period in the repeated game is so small that subjects cannot observe distinct periods, i.e., the reaction
time of the experimental software is lower than the human reaction time. In the context
of oligopoly competition, a continuous time framework thus allows subjects to set and
change prices freely at any time (asynchronous-move), whereas a discrete time setting
requires subjects to decide about prices at fixed points in time (synchronous-move).
This study is not the first to employ a non-discrete time framework and previous comparisons of discrete and continuous time in lab experiments have been conducted for
specific contexts (see Subsection 3.3.1 for an overview). However, it is the first experimental study concerned with the emergence of tacit collusion under oligopolistic competition in continuous time and the first to systematically investigate the differences
in outcomes between continuous and discrete time oligopoly experiments. In summary, the findings suggest, irrespective of the underlying competition model (Bertrand
or Cournot) and the number of firms (two or three), that firms can coordinate better
on collusive outcomes in discrete time oligopoly experiments than in continuous time
oligopoly experiments. This is in sharp contrast to the experimental study by Friedman
and Oprea (2012), who find higher levels of coordination in continuous time (repeated)
than discrete time (one-shot) prisoner’s dilemma games. In conclusion, the nature of
the game seems to be particularly relevant for whether a continuous or discrete time
setting facilitates cooperative behavior.
The remainder of this section is organized as follows. In Subsection 3.3.1, the extant
literature on (near-)continuous time experiments is reviewed and a framework to classify different experimental time frameworks is offered. Subsection 3.3.2 describes the
design and procedures of the conducted experiment in detail. Empirical results of the
78
3.3 Oligopoly competition in continuous time
laboratory experiment are derived in Subsection 3.3.3. Finally, Subsection 3.3.4 discusses methodological and policy implications of the experimental findings.
3.3.1 Timing in experiments
Economic laboratory experiments are used to evaluate theoretical predictions or to assess the implications of economic market designs prior to their application in the field.
For this purpose, human participants face repeated decisions in a given experimental
scenario. In experiments that study competition between firms, repetition is usually implemented as a (fixed or random) number of successive (and otherwise independent)
periods. A period does not start before all subjects have made a decision in the previous
period. This yields synchronous (simultaneous) decision-making by the firms, which
however does not resemble most strategic interactions in reality such as competition
between firms in a market. Instead, firms may make decisions about their products
and prices at any given time and respond to their rivals’ actions accordingly, i.e., decisions are asynchronous. In consequence, experimental economists may often apply
a discrete time framework to model situations in which decisions are actually made in
continuous time.
Since the computerization of economic lab experiments, researchers have implemented
different timing schemes.
However, there is little—although recently growing—
evidence on how decision-making in the lab differs between experimental setups in discrete and continuous time (Berninghaus et al., 2007; Friedman and Oprea, 2012; Oprea
et al., 2014; Kephart and Friedman, 2015; Kephart and Rose, 2015). In lack of a consistent definition, it is unclear which aspects constitute a discrete time framework and
consequently, non-discrete time frameworks in economic lab experiments. Therefore,
a classification of discrete and non-discrete time experiments is proposed, before the
extant literature is reviewed.
79
Chapter 3 Theoretical Foundations and Experimental Methodology
A classification for continuous and discrete time experiments
First, it is defined what is commonly meant by a discrete time experiment: discrete time
is a synchronous-move repeated games framework with an unbounded period length.
A period, i.e., a discrete time step, ends only after all subjects have confirmed their
decisions. All experimental modes that deviate from this set-up are classified as experiments in non-discrete time and are reviewed subsequently. Among the non-discrete
time experiments there exists a variety of modes that can be distinguished further.
The classification of non-discrete time experiments into continuous time and nearcontinuous time experiments is motivated by Freeman and Ambady (2010), who show
that the human reaction time for very simple computerized tasks as measured by the
time needed to process information presented on the screen and to perform a mouse
click is above 0.5 seconds. Thus, in the most conservative way, continuous time experiments are defined as those with rapidly repeated periods (of fixed time length), in
which the length of each period is below the threshold of the human reaction time,
i.e., 0.5 seconds or below. Technically speaking, as computers are designed to perform
operations in discrete steps, a computerized experiment is said to run in continuous
time if the transaction time (period length) between the experimental server and the
clients is smaller than the subjects’ reaction time. In continuous time experiments an
action becomes profit-relevant instantaneously and can be observed by other subjects
accordingly. Thus, the (potential) consequences of an action cannot be tested prior to
making the decision. Moreover, it is virtually impossible for subjects to make decisions simultaneously. Consequently, in continuous time experiments the order or time
of decision-making is not exogenously given and thus, inter-period asynchronous interaction emerges naturally. Subjects can act and react upon each others’ moves at a
self-specified time. Profits and other outcome variables become flow values. Thus, the
key aspect of a continuous time framework is that it endogenizes the timing of decisions
and thereby captures asynchronicity in decision-making as in many real-world strategic interactions, i.e., decision-making that is neither simultaneous nor sequential.
80
3.3 Oligopoly competition in continuous time
Continuous time experiments need to be distinguished from near-continuous time experiments, which employ a synchronous-move repeated games framework with constant,
finite period lengths above the human reaction time, during which subjects have to
decide on their action in the subsequent period. As in continuous time experiments,
individual decisions are transferred from one period to the next, and hence, “doing
nothing” results in choosing the same action as before. Without communication between subjects, decisions by rivals do not become public and profit-relevant before the
end of a period. Therefore, as the reaction time is above the human decision threshold,
interaction is potentially synchronized and decision-making is simultaneous. Hence, as
under discrete time, inter-period asynchronous interaction or even sequential-moving
may occur behaviorally, but not naturally.
The advantages of near-continuous time in comparison to discrete time experiments
are a high control over the length of the session and the possibility to collect a large
amount of data in relatively short time. Thereby, patterns of repeated decisions may
occur that would not have been observable in a discrete time experiment (with fewer
periods). However, this time framework also bears two potential problems. First, different cognitive and physical abilities of human participants may have a greater influence on experimental results than in an experiment run in discrete time, i.e., some
subjects may not be able to change actions fast enough and hence, data on intended
decisions would be lost. Second, the repetition of short periods with a fixed length may
induce an aspiration to “use the time given” and change one’s decision every period.
Both caveats generally apply to continuous time experiments as well. However, since
profit is a flow value in continuous time experiments, a small difference in subjects’ reaction times has only a relatively small impact on profits as subjects can react promptly
to a rival’s decision. For example, in a duopoly the additional profit gained by defecting from a cooperative state is linear in the rival’s reaction time with a continuous
time framework but step-wise constant with a near-continuous time framework. Consequently, for the same rival’s reaction time below the near-continuous period length,
(myopic) profits from defection are higher in near-continuous than in continuous time.
81
Chapter 3 Theoretical Foundations and Experimental Methodology
A potential problem of the continuous time framework is that the theoretical prediction
of the repeated game may change due to its dynamic nature (see Subsection 3.3.2).
Finally, the continuous time framework needs to be distinguished from a clock or deadline mechanism, which is a synchronous-move repeated games framework with constant, finite period lengths under which subjects’ current actions are common knowledge and may be changed (freely), but do not become binding, until a clock runs out
or a deadline is reached. The action chosen in the very last moment before the end of
the deadline is reached becomes binding and constitutes the subject’s profit-relevant
decision for the next period. Consequently, the current action of a subject may be interpreted as an intention for the final decision in the period but is profit-irrelevant, and
thus, cheap talk. Subjects can react to each others’ actions during a period, which can
be referred to as intra-period asynchronous interaction. As Roth (1995) points out, this
experimental design gives some indication of how “last-minute agreements” (p.324)
in negotiations evolve.2 With respect to experimental design, the clock or deadline
mechanism is a hybrid of the continuous time framework and the near-continuous time
framework. Whereas intra-period interaction between subjects (i.e., cheap talk before
the deadline) is asynchronous, inter-period interaction between subjects (i.e., decisionmaking at the deadline) is synchronized. See Roth (1995) for an overview on the effects
of the clock or deadline mechanism and proposed models to explain these effects.
Review of non-discrete time experiments
Table 3.1 lists non-discrete time experiments in the extant literature and classifies them
according to the definitions given above. Next to the type of game that was run in the
laboratory, the length of a period, the mode of asynchronous interaction (i.e., between
2 There
are two further strands of experiments that implement a variant of this clock or deadline mechanism. The first strand (Dorsey, 1992; Goren et al., 2003; Ishii and Kurzban, 2008) introduces restrictions
on how actions may be adjusted during the period, e.g., individual contributions in a public good game
may only be increased but not decreased over time. Kurzban et al. (2001) compares public good experiments in a clock framework with and without revocable contributions. In the second strand (Levati
and Neugebauer, 2004; Murphy et al., 2006), prior to a clock running out, the period may end by other
means such as a player dropping out of an auction or exiting a market. Both strands may be viewed as
extensions to the basic clock/deadline mechanism.
82
3.3 Oligopoly competition in continuous time
or within a period) and the classification to one of the non-discrete time frameworks
are reported for each experimental study. In the following, a period in the context of
repeated games is denoted as the amount of time a subject has to decide on a binding
action. Note that this is identical to the minimum amount of time that a binding decision by a subject holds. Thus, a supergame is defined as a complete sequence of a fixed
or random number of periods.
Feeley et al. (1997), Berninghaus and Ehrhart (2003), and Berninghaus et al. (1999, 2006,
2007) were among the first to conduct continuous time experiments with period lengths
below the human reaction time and a fixed supergame length of several minutes up to
half an hour.3 More recently, Cheung and Friedman (2009), Friedman and Oprea (2012),
Oprea et al. (2014), Bigoni et al. (2015), Kephart and Friedman (2015), and Kephart
and Rose (2015) ran experiments in continuous time with supergame lengths from 20
seconds to four minutes. Berninghaus and Ehrhart (1998), Deck and Wilson (2002, 2003,
2008), Davis (2009), Davis and Korenok (2009), Davis et al. (2009, 2010), and Friedman
et al. (2015) conduct near-continuous time experiments with a high number of rapidly
repeated periods. The clock or deadline framework is employed by Roth et al. (1988),
Güth et al. (2002), Goren et al. (2004), and Deck and Nikiforakis (2012).
Of the continuous time experiments, only Berninghaus et al. (2007), Friedman and
Oprea (2012), Oprea et al. (2014), Kephart and Friedman (2015), and Kephart and Rose
(2015) compare outcomes under both discrete and continuous time. Berninghaus et al.
(2007) study network formation and network effects in social and economic networks
in which connections to other players are beneficial but costly. They find that the formation of a certain star structure, which is the unique Nash equilibrium, prevails under
both time frameworks. However, subjects are found to alternate the coveted position
of the center player in the star network in continuous time but not in discrete time.
Berninghaus et al. (2007) suggest that their results may be explained by inequity aversion. As the discrete treatment is composed of only 15 periods whereas the continuous
3 Note that Millner et al. (1990) already followed a continuous time approach with output variables given
as flow values. However, technical constraints of the PLATO software used for the computerization of
the experiment resulted in a transaction time between clients and server of about five seconds, which
lies one order of magnitude above the suggested threshold of 0.5 seconds.
83
Chapter 3 Theoretical Foundations and Experimental Methodology
TABLE 3.1: Economic laboratory experiments in non-discrete time.
Study
Type of game
Feeley et al. (1997)
Berninghaus et al. (1999)
Berninghaus and Ehrhart (2003)
Berninghaus et al. (2006)
Berninghaus et al. (2007)
Cheung and Friedman (2009)
Knigge and Buskens (2010)
Friedman and Oprea (2012)
Oprea et al. (2014)
Bigoni et al. (2015)
Kephart and Friedman (2015)
Kephart and Rose (2015)
Chapter 5
Prisoner’s dilemma
Population
Evolutionary
Network formation
Network formation
Coordination
Network formation
Prisoner’s dilemma
Public good
Prisoner’s dilemma
Hotelling
Hotelling
Wholesale competition
Period length†
Async. interaction
n/a††
1/10 seconds
1/10 seconds
1/10 seconds
1/5 seconds
1/2 seconds
n/a††
1/20 seconds
1/10 seconds
16/100 seconds
1/20 seconds
1/20 seconds
1/2 seconds
Inter-period
Inter-period
Inter-period
Inter-period
Inter-period
Inter-period
Inter-period
Inter-period
Inter-period
Inter-period
Inter-period
Inter-period
Inter-period
Continuous time
Near-continuous time
(1990)†††
Millner et al.
Berninghaus and Ehrhart (1998)
Deck and Wilson (2002)
Deck and Wilson (2003)
Deck and Wilson (2008)
Davis (2009)
Davis and Korenok (2009)
Davis et al. (2009)
Davis et al. (2010)
Friedman et al. (2015)
Posted offer
Public good
Posted offer
Posted offer
Posted offer
Posted offer
Posted offer
Posted offer
Posted offer
Cournot competition
Roth et al. (1988)
Dorsey (1992)
Kurzban et al. (2001)
Güth et al. (2002)
Goren et al. (2003)
Goren et al. (2004)
Levati and Neugebauer (2004)
Murphy et al. (2006)
Ishii and Kurzban (2008)
Deck and Nikiforakis (2012)
Bargaining
Public good
Public good
Public good
Public good
Public good
Public good
Trust dilemma
Public good
Minimum-effort
5 seconds
10–90 seconds
3 seconds
3 seconds
1.7 seconds
7 seconds
7–70 seconds
12 seconds
12–18 seconds
4 seconds
Inter-period
Inter-period
Inter-k-periods-block
Inter-20-periods-block
Inter-period
Inter-period
Inter-period
Inter-period
Inter-period
Inter-period
Clock/deadline mechanism
†
9–12 minutes
180 seconds
90 seconds
3 minutes
60–90 seconds
60–90 seconds
 50 seconds
 45 seconds
90 seconds
60 seconds
Intra-period
Intra-period
Intra-period
Intra-period
Intra-period
Intra-period
Intra-period
Intra-period
Intra-period
Intra-period
Period length is defined as the minimum time that a binding decision by a subject holds.
The transaction time of the software is not stated, but assumed to be below 0.5 seconds.
††† The experiment uses the PLATO software. Period length is determined as its estimated latency.
††
84
3.3 Oligopoly competition in continuous time
treatment runs for 30 minutes, subjects may find it easier to equalize payoffs among
themselves in the latter.
Recently, Oprea et al. (2014) compared contributions to a public good in discrete time
and continuous time over ten minutes. In the continuous time treatments, contributions
could be changed in real-time. In the discrete time treatments, subjects decided on their
contributions once a minute, i.e., they played ten periods with a fixed length of one
minute. In this setup with few discrete periods (of a limited period length), the authors
find no differences in contributions between the two time frameworks.
By contrast, Friedman and Oprea (2012) compare cooperative behavior in the prisoner’s
dilemma in discrete and continuous time and find that the continuous time framework
fosters cooperation among the players relative to discrete time. More precisely, the
authors compare continuous and discrete variants of the prisoner’s dilemma in supergames with a constant length of 60 seconds. In continuous time, they find a median
mutual cooperation rate of 90 percent over the supergames’ duration. With the duration of each supergame being fixed, the number of periods is decreased to eight in 60
seconds and finally to one in 60 seconds, i.e., a one-shot game.4 The main finding of
the study is that cooperation decreases as the number of periods decreases so that the
median rate of mutual cooperation is zero in the one-shot treatments. In other words,
cooperation is higher in a continuously repeated prisoner’s dilemma than in a one-shot
(discrete) prisoner’s dilemma. Friedman and Oprea analyze the subjects’ individual
behavior in the continuous time treatments and identify different strategies. A model
of e-equilibria (Radner, 1986; Bergin and MacLeod, 1993) is found to predict their findings very well. A key aspect of their experimental design is “that period lengths and
potential payoffs are kept constant across [...] treatments” (Friedman and Oprea, 2012,
p.343). However, this is only achieved by implicitly changing two treatment variables
in the transition from continuous time to the one-shot (discrete) treatment at the same
time. The first treatment variable is obviously the time framework of a repeated game,
4 Comparably,
Berninghaus and Ehrhart (1998) varied the number of periods (10, 30, and 90) in a public
good game of a total fixed session length of 15 minutes and found that cooperation increases with the
number of repetitions.
85
Chapter 3 Theoretical Foundations and Experimental Methodology
i.e., continuous or discrete, and the second treatment variable is the repetition of the
game itself, i.e., repeated game or one-shot game.
Both Kephart and Friedman (2015) as well as Kephart and Rose (2015) compare a discrete time and two continuous time variants of the Hotelling (1929) spatial competition
model with and without vertical differentiation. Kephart and Friedman (2015) find
that under continuous time location choices resemble the static Nash equilibrium more
closely than under discrete time. With vertical differentiation and an additional choice
on price, Kephart and Rose (2015) find some support for the notion that continuous
time increases cooperation. Whereas subjects may decide instantaneously in one of the
continuous time treatments, they may change their decision only gradually at a specified “speed” in the other continuous treatment. However, under discrete time, subjects
have to decide on location (and price in case of Kephart and Rose, 2015) during a three
second time interval. Note that with respect to the timing classifications outlined above,
these discrete time treatments clearly fall under the near-continuous time framework.
3.3.2 Experimental framework
The following experiment studies the impact of continuous time relative to discrete
time on the outcome of experimental oligopoly competition. In an attempt to foster
the robustness of findings, symmetric differentiated Bertrand as well as Cournot competition in duopolies and triopolies each are considered. Thereby, the experiment examines three dichotomous treatment variables (discrete vs. continuous time, Bertrand
vs. Cournot competition, duopolies vs. triopolies) in a full-factorial design, resulting
in a total of eight different treatments. The labels used to refer to the treatments in the
following sections are stated in Table 3.2 by appending abbreviations from left to right,
e.g., RB3 refers to the continuous (real-time) Bertrand triopoly treatment.
86
3.3 Oligopoly competition in continuous time
TABLE 3.2: Treatment variables and their values.
Time framework
Competition model
Number of firms
Discrete time (D)
Continuous time (real-time) (R)
Bertrand (B)
Cournot (C)
Duopoly (2)
Triopoly (3)
Oligopoly competition
Price competition à la Bertrand and quantity competition à la Cournot are the two
workhorse models of IO. When comparing different designs in experiments on firm
behavior, they serve as good proxies for a large share of models on oligopoly competition. As homogeneous price competition yields a discontinuous demand function and
the empirically observed Bertrand paradox is often deemed unrealistic, the model by
Singh and Vives (1984)—which generalizes the Hotelling (1929) model to exploit the
duality between price and quantity competition in differentiated goods—is utilized for
the experiment. More precisely the model’s generalization (Häckner, 2000) to more
than two firms is employed (see, e.g., Suetens and Potters, 2007, for a previous application in the context of oligopoly experiments).
Consider a market with n 2 N firms. Each firm k 2 {1, ..., n} produces a single good.
The firms’ goods are differentiated horizontally but are homogeneous in vertical quality and have identical demand elasticity. Thus, firms are assumed to be symmetric.
Note that asymmetric (inverse) demand may result in additional behavioral effects in
the experiment which are not in focus here, but are further analyzed in Chapter 6 in an
asymmetric application of the model. In the following, a brief sketch of the propensities of the model is provided, while a detailed analysis of the theoretical predictions is
relegated to Appendix A.1.
For the Cournot treatments of quantity competition, the inverse demand for firm k is
given by
pk = w
l qk + q  q j
j6=k
!
87
Chapter 3 Theoretical Foundations and Experimental Methodology
with w, l > 0 and the degree of substitutability q 2 [ 1, 1]. If q < 0 goods are complements, if q = 0 goods are independent of one another, and if q = 1 they are perfect
substitutes. For non-perfect substitutes (q < 1), the corresponding demand function for
firm k in the Bertrand treatments is given by
qk = G
Lpk + Q
 j6=k p j
n 1
with
w
,
l(1 + q (n 1))
1 + q ( n 2)
L=
,
l(1 q )(1 + q (n 1))
q ( n 1)
Q=
,
l(1 q )(1 + q (n 1))
G=
and n as the number of firms with non-negative demand, i.e., firms that have not exited
the market due to a too high price. If qk < 0 firm k exits the market, its quantity is set to
zero, and n is decreased by one. Normalizing costs to zero, firm k’s profit is Pk = pk qk .
For the subsequent empirical analysis, three benchmark outcomes for Bertrand and
Cournot competition are considered, respectively. Note that, although goods are differentiated, equilibrium prices and quantities are the same for all firms. First, under the
Walrasian (competitive) equilibrium firms are assumed to be price-takers so that they
maximize their profit irrespective of their rivals’ decision. Second, the Nash equilibrium assumes that firms choose a price (quantity) such as to maximize their own profit
given their rivals’ prices (quantities). Third, under the collusive outcome firms are assumed to cooperate and hence, employ joint profit maximization (JPM) acting as a single monopolist. It is straightforward to show that P JPM
Nash
PCournot
Nash
P Bertrand
PWalras
for all valid parameter combinations. If goods are substitutes (g > 0), which will be assumed subsequently, Nash prices and profits are higher under Cournot competition
than under Bertrand competition.
88
3.3 Oligopoly competition in continuous time
TABLE 3.3: Scaled theoretical benchmarks of oligopoly competition for each treatment as displayed in the experiment.
Bertrand
p
Walras
q
Walras
P
Walras
p
Duopoly
Nash
=0
= 60.00
=0
= 25.00
q Nash = 45.00
P Nash = 1406.25
p
=0
q
Walras
= 100.00
P
Walras
=0
p
Nash
= 37.50
q Nash = 62.50
P Nash = 1406.25
p JPM = 50.00
p JPM = 50.00
q JPM = 30.00
q JPM = 50.00
P
JPM
= 1875.00
P JPM = 1500.00
pWalras = 0
pWalras = 0
qWalras = 42.86
qWalras = 100.00
P
Triopoly
Cournot
Walras
Walras
=0
PWalras = 0
p Nash = 16.67
p Nash = 30.00
q Nash = 35.71
q Nash = 70.00
P Nash = 1406.25
P Nash = 1406.25
= 50.00
p JPM = 50.00
q JPM = 21.43
q JPM = 50.00
p
JPM
P JPM = 2531.42
P JPM = 1674.22
For the experiment, the parameters of the oligopoly competition model are w = 100,
l = 1, and q = 23 . Consequently, G =
300
2n+1 ,
L=
6n 3
2n+1 ,
and Q =
6n 6
2n+1 .
Table 3.3 shows the
corresponding scaled theoretical benchmarks of the one-shot game for each treatment
as displayed in the experiment.
In a further effort to maximize comparability between treatments and to prevent any
source for behavioral effects other than the treatment, input and output variables are
scaled in the following way. The action space of prices in Bertrand treatments and
quantities in Cournot treatments is equally set to [0, 100] with a minimum increment of
one and the joint profit maximizing action at a price or quantity of 50. This ensures that
the collusive action is not more or less “behaviorally attractive” across treatments and
that the search costs of finding the collusive action are the same in all treatments. With
a similar intention profits are scaled so that they would be equal in Nash equilibrium.
Thereby, a subject playing the Nash equilibrium of the one-shot game—given that its
89
Chapter 3 Theoretical Foundations and Experimental Methodology
competitors play Nash as well—would make identical profits in all treatments. Altogether, this precludes confounding effects of the experimental design and parametrization. Furthermore, perfect information was ensured in all treatments, i.e., subjects got
individual feedback about each competitor’s price, quantity and profit.
Repeated games in discrete time and continuous time
Moving from the one-shot game introduced above to the repeated game implemented
in the experiment, several experimental design implications are inferred from the extant literature. First, in contrast to Friedman and Oprea (2012), repeated games are
employed in the discrete time treatments as well as in the continuous time treatments.
Second, the discrete time treatments are composed of 60 periods—much more than
in Berninghaus et al. (2007) or Oprea et al. (2014)—to reduce differences to continuous time solely due to a longer time horizon of the experiment (Friedman et al., 2015;
Oechssler et al., 2016). Third and contrasting Kephart and Friedman (2015) as well as
Kephart and Rose (2015), in the discrete time treatments the time provided to subjects
for their decision-making process in each period is not limited. Fourth, discrete time
sessions were ran first and the duration of the continuous time sessions was then set to
equal the average duration of the discrete time sessions, which was about 30 minutes.
Hence, the total session length is similar across all treatments and one period in discrete time corresponds to 30 seconds in continuous time. The period length in the discrete time treatments is infinite and 0.2 seconds in the continuous time treatments, i.e.,
considerably below the conservative threshold of 0.5 seconds. Under the continuous
time framework, current profit represents a flow value of time. In order to maximize
comparability between treatments, the profit displayed in the experimental software
is thus scaled to the profit that subjects would earn if the current prices or quantities
would be held constant for 30 seconds, ceteris paribus. Thereby, given the same prices
or quantities in one of the discrete time treatments and the corresponding continuous
time treatment, the information presented to the subjects is not only qualitatively equal
but also visually identical.
90
3.3 Oligopoly competition in continuous time
The model of differentiated Bertrand and Cournot competition considered in this experiment has a unique strict Nash equilibrium in the one-shot (stage) game. In discrete
time, this also constitutes the unique subgame perfect equilibrium of the finitely repeated game. In continuous time, however, the theoretical prediction is not straightforward. Maskin and Tirole (1988a,b) consider two different continuous time frameworks
with endogenous timing in duopolistic price and quantity competition and show that
equilibrium behavior is similar to a sequential-move infinitely repeated duopoly. In
particular, continuous time is modeled as a fine grid of periods in a sequential-move
game, where firms are committed to a price or quantity for a deterministic or stochastic length of time. Whereas the deterministic variant may rather apply to repeated
games that were classified as near-continuous time, the stochastic variant does capture
the asynchronous nature of continuous time quite well. Irrespective of the time variant, a collusive equilibrium emerges for discount factors close to one. In their model,
Maskin and Tirole assume that the Markov property holds, i.e., that future states of
the stochastic process only depend on the current state and not the sequence of states
that preceded it. In a comparable fashion, Simon and Stinchcombe (1989) model continuous time as “a discrete time model, but with a grid that is infinitely fine” (p.1171)
and thereby suggest a more general definition of games in continuous time. Friedman
and Oprea (2012) point out that this model predicts mutual cooperation at all times in
a prisoner’s dilemma, which may be viewed as a highly abstracted variant of homogeneous Bertrand competition. In sum, theory predicts that, if anything, asynchronousmove continuous time is more prone to tacit collusion than simultaneous-move discrete
time. Additionally, Bigoni et al. (2015) find that a deterministic ending rule facilitates
cooperation even more than a stochastic ending rule under continuous time, whereas
other experimental evidence indicates that the opposite may hold under discrete time
(Dal Bó, 2005). Theses findings add further support to the conjecture that the continuous time treatments in this experiment are expected to exhibit more tacit collusion than
the discrete time experiments.
91
Chapter 3 Theoretical Foundations and Experimental Methodology
Measuring the degree of tacit collusion
As Nash prices, quantities, and profits do not coincide under Bertrand and Cournot
and are additionally dependent on the number of competitors, these values are not adequate to compare cooperative intentions, i.e., tacit collusion, across treatments. This
study therefore combines indices for the degree of tacit collusion used by Engel (2007)
and Suetens and Potters (2007) to compare tacit collusion between treatments irrespective of different theoretical predictions. Thereby, the degree of tacit collusion is measured as the relative deviation of a price, quantity, or profit from the theoretical prediction towards the joint profit maximizing price, quantity, or profit. With respect to
Bertrand (Cournot) competition, a price (quantity) set by a firm can be unambiguously
converted to a degree of tacit collusion. Hence, for means of comparison between treatments, firms may be assumed to decide on a certain degree of tacit collusion instead of
a price or quantity. In a similar fashion, a firm’s profit as well as average profit of firms
in a market may be expressed as a degree of tacit collusion. Therefore, a degree of tacit
collusion based on model input, i.e., price in Bertrand and quantity in Cournot, as well
as a degree of tacit collusion based on model output, i.e., profit is considered. Formally,
the degree of tacit collusion is
j Ex =
x
x JPM
xE
,
xE
with x 2 { p/q, P} and E 2 { Nash, Walras}, resulting in four different measures depending on the theoretical benchmark (Nash or Walrasian equilibrium) and on the in-
put or output variable (price/quantity or profit). Figure 3.2 illustrates the degree of
tacit collusion based on prices as measured relative to the Nash equilibrium, given the
observed price p. If j Ex = 0, the value of x corresponds to the theoretical prediction by
the equilibrium concept E. If j Ex = 1, the market is completely collusive and competitors behave as a single monopolist. Note that j Ep/q may exceed one as joint profit is not
E  1.
monotonic in price or quantity, but jP
92
3.3 Oligopoly competition in continuous time
F IGURE 3.2: Normalized degree of tacit collusion based on prices relative to the Nash equilibrium (based on Engel, 2007, p.494).
pJPM - pNash
p - pNash
pWalras
pNash
p
pJPM
Experimental procedures
With respect to the technical requirements of continuous time and to ensure high control over the correct scaling of time in all treatments, the experiment was computerized with Brownie, a newly-developed Java-based experimental software (Müller et al.,
2014).5 All sessions were run at the Karlsruhe Institute of Technology in Karlsruhe,
Germany between October and December 2014. First the four discrete time treatments
with 60 periods were ran. Without the first period, in which subjects familiarized themselves with the experimental software and decided on their initial price or quantity, the
discrete time sessions took on average roughly 30 minutes. This amount of time was
used to parametrize the length (again, without the phase of deciding on initial price
or quantity) of the subsequently run continuous time sessions. Note that no practice
periods with interaction between subjects were run and thus, no learning confounds
may occur. The matching of subjects was constant throughout a session (fixed partner matching). In total, 240 students of economic fields participated in the experiment.
Subjects were recruited via the ORSEE platform (Greiner, 2015) and participated only
in one of the treatments (between-subject design).
The protocol for each session follows five steps. First, upon entering the lab, subjects
are randomly assigned to a chair, from which they can neither see nor speak to any
other participant of the experiment. Second, after everyone has been seated, the experimental instructions are handed out to the participants in print and read aloud from
5 Recently,
further experimental software that captures continuous time was introduced by Pettit et al.
(2014), particularly for experimenters with limited programming skills, and by Hawkins (2015) for
web-based experiments.
93
Chapter 3 Theoretical Foundations and Experimental Methodology
a recording.6 The recording ensures that any confounding effect of the reader’s voice,
accent, or intonation is identical across sessions from the same treatment and as similar as possible across treatments. Therefore, identical paragraphs across treatments are
recorded once and the recording is used in all treatments. Third, prior to the beginning
of the experiment, each participant has to complete a computerized test of questions
regarding the comprehension of the instructions. It is only allowed to proceed to the
next question after the correct answer to the current question is entered. Fourth, after
all subjects have successfully completed the test, the experiment starts automatically.
Over the course of the experiment participants wear ear protectors so that they are not
influenced by clicking noises of computer mouses or other disturbing noise. Fifth, following the end of the experiment, each participant is paid out the profits accumulated
during the experiment privately and in cash. Following this protocol, the total length
of a session from subjects’ entering to leaving the lab was about one hour. The average
payoff per subject was EUR 16.85.
3.3.3 Results
The experimental data amounts to 12 independent duopolies or triopolies in each treatment. Due to no-shows, two exceptions are the RB3 treatment for which there are only
11 triopolies and the RC3 treatment for which there is data on 13 triopolies, because
the number of no-shows required an additional session. For each cohort there is data
on market variables over 60 periods in a discrete time treatment and on 9,000 ticks (at
an interval of 0.2 seconds each) for each market in a continuous time treatment. In
the following the experimental data is first analyzed on the level of independent cohorts, followed by a panel analysis that considers observations at the tick and period
level. Table 3.4 reports degrees of tacit collusion across treatments averaged over cohorts and gives a first impression on treatment effects.7 Most notably, in stark contrast
6 As
an example, the experimental instructions of the RB3 treatment together with a screenshot of the
experimental software are provided in Appendix C.1.
7 To reduce distortions by start- and end-game effects, the first and last sixth of periods have been dropped
before computing cohort averages. For comparison to Table 3.4, average degrees of tacit collusion over
the entire time horizon across treatments are reported in Appendix B.1.
94
3.3 Oligopoly competition in continuous time
TABLE 3.4: Average degrees of tacit collusion across treatments.
Treatment
Obs.
j̄ Nash
p/q
Nash
j̄P
j̄Walras
p/q
j̄Walras
P
DB2
12
0.860
(0.285)
0.861
(0.326)
0.930
(0.142)
0.965
(0.081)
DB3
12
0.659
(0.352)
0.683
(0.327)
0.773
(0.235)
0.859
(0.145)
DC2
12
0.674
(0.574)
0.532
(0.994)
0.918
(0.143)
0.971
(0.062)
DC3
12
0.473
(0.551)
0.364
(0.785)
0.789
(0.220)
0.898
(0.126)
RB2
12
0.769
(0.371)
0.736
(0.468)
0.885
(0.185)
0.934
(0.117)
RB3
11
0.555
(0.329)
0.505
(0.350)
0.703
(0.219)
0.780
(0.156)
RC2
12
0.842
(0.279)
0.760
(0.374)
0.960
(0.070)
0.985
(0.023)
RC3
13
0.424
0.233
(0.516)
(0.784)
Standard deviations in parentheses.
0.770
(0.206)
0.877
(0.125)
to the hypothesis, the degree of tacit collusion based on Nash profits is significantly
higher under discrete time than under continuous time in a non-parametric one-tailed
Mann-Whitney U test without controlling for the competition model or the number of
competitors (z = 1.77, p = 0.038).
In order to allow for a comparison of panel data from both time frameworks, the experimental data from the continuous time treatments is mapped to the 60 periods of the
discrete time treatments. In particular, for each discrete period, the degree of tacit collusion in the continuous time treatments is averaged over 30 seconds, i.e., 150 consecutive
ticks of 0.2 seconds. Thereby, the first 30 seconds correspond to the first discrete period,
the next 30 seconds correspond to the second discrete period, and so on. The mean is
used as a single proxy for the behavior over 30 seconds as it has the advantage that
a maximum of information about the distribution is preserved and that it is, loosely
speaking, merely a reduction in data resolution rather than a reduction in data itself. In
contrast to the median or other point statistics, changing the value of any single data
95
Chapter 3 Theoretical Foundations and Experimental Methodology
point inevitably changes the mean as well. For a direct comparison of the two time
frameworks using the mean is therefore arguably most conservative.
R ESULT 3.1. The degree of tacit collusion based on profits is significantly higher under discrete
time than under continuous time.
A firm’s profit is determined not only by its own decisions but also by the decisions of
its rivals. One firm’s profit in a period is hence not independent from its rivals’ profits. Therefore, the degree of tacit collusion based on profits is measured on the market
level, i.e., calculating the average of each firm’s profit in a duopoly or triopoly. There
are a total of 96 markets across all treatments with 60 discretized periods each. Testing for treatment effects in panel data requires to control for the dependence between
observations from the same market as opposed to observations from different markets.
Consequently, the following multilevel mixed-effects regression model is estimated, for
which treatment DB2 serves as a baseline:
E
jP,t,m
= b0 + x m
+ b Continuous · Continuous
+ b Cournot · Cournot
+ b Triopoly · Triopoly
+ ( b Period + b Period,m ) · t
+ et,m ,
E
where jP,t,m
is the degree of tacit collusion based on average profit P of all firms on
market, i.e., duopoly or triopoly, m in period t. On the market level, x m is the random
intercept that controls for intra-cluster correlation in terms of different base levels of
tacit collusion between markets and b Period,m is a random coefficient for the time trend
in each market.
Table 3.5 reports estimates for the degree of tacit collusion based on Nash (Model 1)
and Walrasian (Model 2) profits, respectively. Irrespective of the theoretical benchmark,
96
3.3 Oligopoly competition in continuous time
continuous time is found to have a significant negative effect on tacit collusion and reduces the degree of tacit collusion between 4 and 20 percentage points (pp), ceteris
paribus. This is in stark contrast to the hypothesis and previous experimental findings. Yet, both control treatment dummies for the competition model and the number
of firms show the expected effects. In line with the meta-study on number effects in
oligopoly experiments in Chapter 6, triopolies exhibit (10 to 22 pp) less tacit collusion
than duopolies. Moreover, price competition is found to facilitate tacit collusion compared to quantity competition if measured based on Nash profit. The degree of tacit
collusion is almost 26 pp lower under quantity competition compared to price competition. However, the finding is reversed if tacit collusion is measured based on Walrasian
profit. Then, the degree of tacit collusion under quantity competition lies almost 5 pp
above price competition, everything else being equal. This is also in line with the expectation as the Walrasian equilibrium is independent of the competition model so that
the Walrasian-based degree of tacit collusion does not control for the different Nash
predictions of price and quantity competition. Furthermore, there are no significant interaction effects between the treatment variables in either regression model. In sum, the
effect of continuous time compared to discrete time is not only statistically significant
but similar in magnitude to the effects due to the number of competitors and the mode
of competition.
In addition to the assessment of collusion degrees based on profits, i.e., an output variable, a similar yet complementary analysis of prices and quantities, i.e., input variables,
is conducted next. Thereby, instead of aggregate market behavior, the individual firm
choices of prices and quantities are compared across treatments.
R ESULT 3.2. The degree of tacit collusion based on prices and quantities is significantly higher
under discrete time than under continuous time.
Each decision by a firm on a price or quantity can be unambiguously transferred into
a choice for a certain degree of tacit collusion, which makes decisions on prices and
quantities comparable across treatments. Applying the same approach as above, the
97
Chapter 3 Theoretical Foundations and Experimental Methodology
TABLE 3.5: Multilevel mixed-effects linear regressions of the degree of tacit collusion on treatments.
Baseline: Discrete Bertrand Duopoly (DB2).
Covariate
(1)
Nash
jP
(2)
jWalras
P
(3)
j Nash
p/q
(4)
jWalras
p/q
(5)
j Nash
p/q  1
(6)
jWalras
1
p/q
Continuous
0.196⇤
(0.110)
0.037⇤
(0.022)
0.109⇤⇤
(0.052)
0.092⇤⇤
(0.037)
0.106⇤⇤
(0.053)
0.094⇤⇤
(0.038)
Cournot
0.264⇤⇤
(0.110)
0.048⇤⇤
(0.022)
0.276⇤⇤⇤
(0.052)
0.053
(0.037)
0.235⇤⇤⇤
(0.053)
0.003
(0.038)
Triopoly
0.219⇤⇤
(0.110)
0.097⇤⇤⇤
(0.022)
0.179⇤⇤⇤
(0.054)
0.138⇤⇤⇤
(0.038)
0.190⇤⇤⇤
(0.054)
0.145⇤⇤⇤
(0.039)
Period
< 0.001
(0.002)
> 0.001
(< 0.001)
Constant
0.851⇤⇤⇤
(0.110)
0.941⇤⇤⇤
(0.022)
Groups
Observations
96
5,760
96
5,760
>
0.001
(0.001)
0.001
(< 0.001)
0.914⇤⇤⇤
(0.055)
240
14,400
0.883⇤⇤⇤
(0.039)
240
14,400
< 0.001
(0.001)
0.853⇤⇤⇤
(0.056)
240
13,876
< 0.001
(< 0.001)
0.823⇤⇤⇤
(0.040)
240
13,876
Standard errors in parentheses.
⇤ p < 0.10, ⇤⇤ p < 0.05, ⇤⇤⇤ p < 0.01
following multilevel mixed-effects regression model is used to estimate firms’ behavior
over time as measured by the degree of tacit collusion and to control for firm-specific
random effects:
j Ep/q,t,k = b 0 + x k
+ b Continuous · Continuous
+ b Cournot · Cournot
+ b Triopoly · Triopoly
+ ( b Period + b Period,k ) · t
+ et,k ,
with j Ep/q,t,k as the degree of tacit collusion based on firm k’s price p or quantity q played
in period t.
Estimation results for Models 3 and 4, reported in Table 3.5, confirm that the degree of
tacit collusion of firms’ actions is significantly higher (9 to 11 pp) under discrete time
than under continuous time, both with respect to Nash equilibrium as well as Wal-
98
3.3 Oligopoly competition in continuous time
rasian equilibrium. Similarly, prices and quantities in triopolies are 14 to 18 pp less
collusive than in duopolies. With respect to the mode of competition, however, price
competition elicits more collusive behavior than quantity competition irrespective of
the underlying theoretical benchmark. Although the difference is found to be significant and economically relevant with almost 28 pp only in the Nash-based degree of tacit
collusion, quantity competition is—contrary to expectations—not more prone to tacit
collusion based on Walrasian equilibrium. Again, there are no significant interaction
effects between treatment variables.
A possible criticism of the previous analysis is that j Ep/q is not monotonic in “collusiveness” as it may exceed one, although firms make lower profit compared to the case of
j Ep/q = 1. In fact, 3.6% of firms’ prices and quantities exceed x JPM . However, any value
of j Ep/q < 1 is a deviation from the collusive equilibrium. This is not captured in Models 3 and 4 in Table 3.5. Excluding all observations with j Ep/q > 1 results in estimates
reported in Models 5 and 6, which show that the treatment effects are robust to degrees
of tacit collusion exceeding one. Other alternatives dealing with these outliers such as
folding down all observations with a tacit collusion degree above one, i.e., choosing
1
1
j Ep/q as the dependent variable, lead to similar results.
3.3.4 Discussion
This study provides empirical evidence that tacit collusion is higher in discrete time
experimental oligopolies than in continuous time experimental oligopolies. Thereby,
discrete time is based on synchronized and simultaneous decision-making and continuous time is based on asynchronous and endogenized decision-making. These findings are robust with respect to a full-factorial treatment design with i) differentiated
Bertrand and Cournot competition, ii) in duopolies and triopolies, iii) under discrete
time and continuous time. The key insights can be summarized as follows: First, controlling for the competition model as well as the number of firms there is significantly
more tacit collusion under discrete time than under continuous time, irrespective of
99
Chapter 3 Theoretical Foundations and Experimental Methodology
whether the Nash or the Walrasian equilibrium serves as the relative benchmark. This
is in stark contrast to the theory (Maskin and Tirole, 1988a,b; Simon and Stinchcombe,
1989) as well as previous experimental studies on continuous and discrete time (Friedman and Oprea, 2012; Oprea et al., 2014). Second, the study confirms two well-known
relationships obtained in previous studies: Duopolies are found to be more collusive
than triopolies (see Chapter 6); and Bertrand competition in prices is found to be more
prone to tacit collusion than Cournot competition in quantities (Suetens and Potters,
2007). These latter findings indicate that participants in the experiment behave in line
with the general theory on oligopoly competition and thus findings with respect to the
effect of the time framework cannot be simply dismissed as experimental artifacts.
In this vein, the implications for further research are two-fold. First, researchers designing oligopoly experiments should consider that the implementation of a certain
time framework may alter their results. In particular, experiments on tacit collusion—
which were until now solely run in discrete time—may have potentially overestimated
the supra-competitive effect. Furthermore, it cannot be ruled out that the mode of timing interacts with other properties of oligopoly competition such as market demand,
cost structure or strategy space. Second and more general, the effect of continuous
time on repeated non-cooperative games is ambiguous. In contrast to this study, experiments on simpler games such as contribution to a public good (Oprea et al., 2014)
or the prisoner’s dilemma (Friedman and Oprea, 2012) found no differences between
time frameworks or even higher propensities to cooperate under continuous time than
under discrete time. The experiment differs from these two studies in several ways,
especially with regard to a greater action space and a higher number of periods in the
discrete time treatments. Thus, it may prove worthwhile to systematically vary the
number of periods in future research on discrete time versus continuous time and extend the comparison to other games (with a different number of possible actions).
For a deeper understanding of why oligopolistic firms find it easier to tacitly collude
under discrete time than under continuous time a more profound analysis of firms’ behavior is required. Firms may apply different strategies or learn from past behavior in
100
3.3 Oligopoly competition in continuous time
many different ways: For example, behavior by firms in repeated oligopoly competition
may be characterized by a static strategy (see, e.g., Fudenberg et al. (2012) for strategies
in a prisoner’s dilemma), by a dynamic strategy such as the imitation of a competitor’s behavior (Huck et al., 1999) or by learning from own and competitors’ decisions
in the past (Huck et al., 2004a). In particular, reinforcement learning, which was previously found to converge to collusion in a homogeneous Cournot oligopoly (Waltman
and Kaymak, 2008), may be a fruitful approach in explicitly capturing the different dynamics of simultaneous-move discrete time and asynchronous-move continuous time.
Furthermore, as continuous time makes simultaneous decision-making virtually impossible, experiments comparing sequential-move and simultaneous-move games may
be connected to the findings of this study. In fact, experiments on quantity (Huck et al.,
2001) and price (Kübler and Müller, 2002) competition suggest that sequential-move
interaction is less prone to tacit collusion than simultaneous-move competition. However, this finding holds only if the sequence of decision-making is exogenous. Instead, if
timing of sequential decisions is endogenous, behavior is equal to simultaneous-move
oligopolies (Fonseca et al., 2005; Müller, 2006).
A key feature of the presented experimental design is that it compares two extremes
of a spectrum of time frameworks with each other: Pure discrete time with no limit
on period lengths and pure continuous time with a fixed period length below the human reaction time. Obviously, this design inherently cedes control over the duration of
sessions in discrete time. Therefore, a further investigation of period lengths in the transition from discrete time to continuous time may provide valuable insights whether the
effect of the time framework is driven by period length, number of repetitions, or the
(a)synchronicity of decision-making. In particular, this calls for an experiment in nearcontinuous time with different period lengths and different numbers of periods.
101
Chapter 4
Infrastructure Competition and Open
Access Regulation
M
ARGIN squeezes can occur in markets where non-integrated downstream
firms, which supply only retail goods, rely on wholesale access to an essential
upstream good provided by a vertically integrated competitor. In this case, the integrated firm may be able to set the wholesale price above the retail price, i.e., to squeeze
the margin of the downstream firm, and thus to ultimately induce its exit from the retail
market, i.e., to foreclose the non-integrated rival. Whether regulators or antitrust authorities should intervene in cases of margin squeeze conduct is controversial. European
agencies and courts qualify margin squeeze conduct as a stand-alone antitrust abuse
(see the cases Deutsche Telekom, Telefónica, and TeliaSonera), whereas US courts dismiss
allegations based on the margin squeeze rationale (see the cases linkLine and Trinko).
Auf’mkolk (2012) reviews recent margin squeeze cases in competition law, Gaudin and
Saavedra (2014) discuss margin squeeze regulation in European telecommunications
markets.
This chapter scrutinizes Open Access regulation that prohibits margin squeezes in markets with more than one integrated firm producing the upstream good. More specifically, a retail triopoly is considered in which one vertically integrated firm provides
This chapter is based on joint work with Niklas Horstmann and Jan Krämer (Horstmann et al., 2016a).
103
Chapter 4 Infrastructure Competition and Open Access Regulation
wholesale access to a non-integrated downstream competitor, while the other integrated firm relies on self-supply of the upstream good and does not offer access. This
generic market structure captures any industry in which downstream firms are supplied by a wholesale monopoly while other integrated firms exercise additional competitive pressure in the retail market. In particular, the scenario of infrastructure competition in conjunction with retail competition resembles the current state of many European telecommunications markets, where former telecommunications incumbents
compete with integrated cable operators as well as with retailers that rely on the incumbent’s access network as an input.
The extant economic literature on margin squeeze regulation focuses on market structures with a single integrated monopolist under different settings. First, starting from a
setting with homogeneous retail goods, Jullien et al. (2014) point to ambiguous effects
of banning margin squeezes: although wholesale prices decrease, retail prices may increase. In other words, non-integrated retailers benefit from margin squeeze regulation, whereas consumers may be worse off due to more severe double marginalization.
Second, Ergas et al. (2010) suggest that retail goods are not likely to be homogeneous
and Petulowa and Saavedra (2014) show that with horizontally and vertically differentiated products the single integrated firm will only engage in a margin squeeze if
its non-integrated competitor is more efficient. In this setting it is found that a margin squeeze ban induces an increase in the integrated firm’s retail price, but nevertheless ultimately benefits consumers, because the retail price of the non-integrated firm
decreases provided that upstream market regulation is non-constraining. Third, in a
market setting that resembles a fixed voice telephony market, Briglauer et al. (2011)
demonstrate that increasing infrastructure competition from non-strategic rivals may
elicit a margin squeeze depending on access price regulation.
In this chapter, the impact of horizontally differentiated retail goods under infrastructure competition with strategic competitors is scrutinized and it is shown that due to
increased competitive pressure in the retail market, margin squeezes occur also in the
case when competitors are equally efficient.
104
Bouckaert and Verboven (2004) identify three types of margin squeezes according to
the prevailing regulatory regime: regulatory price squeezes (i.e., if wholesale and retail prices are regulated), predatory price squeezes (i.e., if only wholesale prices are
regulated), and foreclosure (i.e., if no prices are regulated). Here, the focus is on the latter type because in industries of competing vertically integrated firms margin squeeze
regulation is considered a potential substitute to access price regulation, and not a complement to it. This is exemplified by the ex ante economic replicability test from the European Commission’s (2013a) recommendation on consistent non-discrimination (see
Chapter 2 and Jaunaux and Lebourges, 2015, for a discussion of the recommendation
and the ex ante test, respectively).
Höffler and Schmidt (2008) also consider strategic interaction between infrastructure
operators and retailers and show that, with differentiated retail goods, consumer surplus decreases under retail minus X regulation, which is akin to a margin squeeze ban.
While their market structure is comparable to the setting in this chapter, Höffler and
Schmidt do not consider the emergence of foreclosure, which in fact constitutes an equilibrium outcome as noted by Atiyas et al. (2015). However, as highlighted by Bouckaert
and Verboven (2004) and Gaudin and Mantzari (2016), foreclosure is a central concern
with regard to margin squeeze conduct. Therefore, the following analysis will distinguish between this more severe exclusionary behavior from simple exploitative margin
squeeze conduct (Jullien et al., 2014). The findings show that margin squeeze regulation prevents both foreclosure and margin squeezes, which benefits a non-integrated
competitor, but does not translate into a benefit for consumers. Whereas the effect on
the access provider’s profit depends on product differentiation and price setting in the
retail market, margin squeeze regulation unambiguously increases profit of the integrated firm, which does not supply the wholesale input. On the contrary, it is found
that margin squeeze regulation is always detrimental to consumers in the presence of
infrastructure competition.
The remainder of this chapter is organized as follows. Next, the general market structure and model that is used to analyze the effect of MSR in lieu of no regulation (NR) is
105
Chapter 4 Infrastructure Competition and Open Access Regulation
described. In Sections 4.2 and 4.3 two different timing variants of the model are studied.
Section 4.4 discusses the policy implications and limitations of the presented model.
4.1 Theoretical framework
Consider the industry depicted in Figure 4.1 with two vertically integrated firms
(Firm A & Firm B) and a non-integrated firm (Firm D) which operates only in the
downstream market. For each unit of its retail good Firm D is required to purchase
a unit of the homogeneous upstream good, which Firm A offers at price a. Firm B does
not provide wholesale access. The retail price of firm k 2 { A, B, D } is denoted by pk .
F IGURE 4.1: Market structure (based on Bourreau et al., 2011, p. 683).
Wholesale
market
Firm B
Firm A
Wholesale
good
a
Wholesale
good
Retail
market
Firm D
Retail
good
Retail
good
pA
pD
Retail
good
pB
Consumers
Assuming the representative consumer suggested by Shubik and Levitan (1980) retail
goods are horizontally differentiated and demand for firm k’s retail good is given by
p A + pB + pD
)
3
provided that none of the firms exits the downstream
market (Höffler, 2008). Thereby, g
0 denotes the degree of substitutability so that high
qk = 13 (1
pk
g( pk
values indicate less differentiated retail goods. In contrast to the case of homogeneous
retail goods, competition in differentiated goods requires to distinguish between a margin squeeze and foreclosure of the downstream firm. More specifically, even if Firm A engages in a margin squeeze, i.e., D := p A
a < 0, Firm D may still make a positive profit,
because consumers value variety. If, however, the spread between Firm A’s wholesale
and retail price exceeds a certain threshold that depends on the degree of substitutabil-
106
4.2 Competitive retail fringe
ity, i.e., D < D(g), Firm D cannot make positive profits and therefore exits the market,
i.e., it is effectively foreclosed.
In this vein, margin squeeze regulation serves two objectives: (i) to establish a level
playing field by prescribing that no firm sets its retail price below the market wholesale
price and (ii) to safeguard product variety by ensuring that non-integrated firms are
not foreclosed from the market. Whether it is profitable for the access provider, Firm A,
to foreclose the downstream retailer, Firm D, with respect to the model depends on
the degree of substitutability, g, and the strategic role of Firm D. In Section 4.2 Firm D
is considered to act as a competitive fringe, which sets its retail price after observing
the retail prices of the integrated firms. In the absence of MSR, i.e., under NR, Firm D
will be subjected to a margin squeeze in this scenario, but will never be foreclosed. In
Section 4.3 an alternative timing of the model is considered, where Firm D sets its retail
price simultaneously with the integrated firms. In this scenario, both margin squeeze
and foreclosure may occur under NR.
4.2 Competitive retail fringe
When Firm D is thought of as a competitive fringe that reacts to the integrated firms’
prices, the timing of the model is as follows:
Stage 1: Firm A sets the wholesale price a.
Stage 2: Firm A and Firm B set their respective retail prices p A and p B .
Stage 3: Firm D sets its retail price p D .
The subgame-perfect equilibrium of this game is determined by backward induction for the case of NR, where Firm A can set its prices freely, and for the case of
MSR, where Firm A must adhere to the constraint D
0. For a welfare analysis of
MSR in lieu of NR not only firms’ prices pk , quantities qk , and profits pk are compared, but also producer surplus PS = Âk pk , i.e., the sum of firms’ profits, consumer
107
Chapter 4 Infrastructure Competition and Open Access Regulation
surplus CS = Âk qk
3
( q2
2(1+ g ) Â k k
+ g3 (Âk qk )2 )
Âk qk pk , i.e., the representative con-
sumer’s net utility (Bouckaert and Kort, 2014), and total surplus TS = PS + CS. In
order to assess MSR relative to NR, ratios fXk =
XkMSR
XkNR
are reported, where X is the mar-
ket variable under investigation. In the following, a sketch of the analysis and some
intuition for the results are offered, whereas the technical details are relegated to Appendix A.2.
R ESULT 4.1. If the access provider has no incentive to foreclose its downstream rival under no
regulation, margin squeeze regulation increases wholesale and retail prices, which leads to a loss
in consumer surplus and total surplus.
Under NR, in Stage 3 Firm D’s first order condition is
1 2ag+gp A +gp B +3a+3
2
3+2g
∂p D
∂p D
NR =
= 0, which yields p D
as the best response to the integrated firms’ prices. In stage 2, both
integrated firms i 2 { A, B} solve
NR )
∂pi ( p D
∂pi
NR
= 0 simultaneously, yielding p NR
A and p B ,
respectively. Anticipating these decisions, Firm A solves
NR NR
∂p A ( p NR
A ,p B ,p D )
∂a
= 0 in Stage
1 and sets the optimal wholesale charge a NR accordingly. The left panel of Figure 4.2
depicts the equilibrium retail and wholesale prices. Notice that Firm A violates the margin squeeze condition for all g > 0. Yet, Firm D makes positive profits in equilibrium as
Firm A has no incentive to foreclose the retailer. In other words, Firm A benefits more
from the wholesale revenue effect than it suffers from the business stealing effect. In contrast, Firm B prefers foreclosure of Firm D, because it suffers from the business stealing
effect and has no wholesale revenue to compensate for this. Relatedly, Firm B sets lower
retail prices than Firm A due to the softening effect (Bourreau et al., 2011; Fudenberg and
Tirole, 1984), which occurs because Firm B has no opportunity cost in terms of foregone
wholesale revenue when decreasing its retail price.
Under MSR, however, Firm A’s retail price setting in Stage 2 is constrained, because
a margin squeeze would occur for all g > 0 under NR. This induces Firm A to raise
its retail price to the level of the profit maximizing wholesale price, i.e., p MSR
= a MSR .
A
Otherwise, wholesale revenue, business stealing, and softening effects are qualitatively
108
4.2 Competitive retail fringe
F IGURE 4.2: Equilibrium prices under NR and MSR.
the same as under NR which preserves the relative order of equilibrium prices (see the
right panel of Figure 4.2).
Figure 4.3a depicts the net effect of MSR in comparison to NR with respect to its relative impact on prices, quantities, and profits. First, the reported ratios demonstrate
that all prices rise under MSR, which highlights that the regulation not only fails to exert a negative impact on the wholesale price, but instead allows firms to attain higher
prices in both wholesale and retail markets. Since the margin squeeze condition is
binding, Firm A has no incentive to lower its retail price following an increase in its
wholesale price. This in turn incentivizes Firm B to increase its retail price as well, because downstream prices of the integrated firms are strategic complements. In addition,
Firm D raises its retail price due to the increased wholesale input price. Second, despite
this universal price increase, retail demands for Firm B and Firm D increase, and only
Firm A’s demand decreases due to the relative magnitude of its price surge compared
to NR. Third, profits increase for all firms under MSR. Whereas Firm A’s downstream
profit deteriorates due to a decline in retail demand, the increase in wholesale revenue
ultimately leads to a net benefit for the access provider as shown in Figure 4.3b.
The welfare analysis of MSR in relation to NR reinforces insights on the effect of MSR
gained thus far (see Figure 4.3c). While producer surplus increases under MSR as a
109
110
( B ) Firm A’s total, upstream, and downstream profit.
( C ) Consumer surplus, producer surplus, and total surplus.
( A ) Firms’ prices, quantities, and profits.
F IGURE 4.3: MSR effect on market variables in the scenario with a competitive retail fringe.
Chapter 4 Infrastructure Competition and Open Access Regulation
4.3 Simultaneous retail pricing
direct consequence of firms’ ability to unanimously reap higher profits, consumer surplus is lower than under NR. More specifically, consumers are worse off because the
increased outputs of Firm B and Firm D are outweighed by higher retail prices of all
firms. Remarkably, note that the harm to consumers due to MSR decreases with increasing substitutability of retail goods. Overall, the effects on producer and consumer
surplus amount to an ultimately negative impact of MSR on total surplus, which is
more severe for more differentiated retail goods, i.e., for lower values of g. Altogether
these results suggest that under the given market structure and timing, MSR cannot be
justified by either a consumer welfare perspective nor a total welfare standard.
4.3 Simultaneous retail pricing
The scenario studied in the previous section may be contested based on the claim that
MSR is particularly relevant and potentially more effective in a scenario where the access provider does not only engage in a margin squeeze but is also likely to effectively
foreclose the non-integrated retailer from the downstream market. Thus, a second scenario, where the competitive position of the non-integrated retailer is strengthened,
and thus foreclosure constitutes an equilibrium under NR if retail goods are close substitutes is scrutinized in the following. In particular, the case where all (integrated and
non-integrated) firms choose their retail prices simultaneously, which reduces the previous three-stage game to a two-stage game is now considered. Again, the focus is
on the cornerstones of the analysis, whereas the technical details are provided in Appendix A.2.
R ESULT 4.2. If the access provider has an incentive to foreclose its downstream rival under no
regulation, margin squeeze regulation prevents the market exit of the non-integrated retailer.
Whereas total surplus increases due to higher profits, consumer surplus always decreases.
Under NR, in Stage 2 all firms choose their retail prices given the wholesale price. In
Stage 1, Firm A decides on its wholesale price by trading off its profits when it does or
111
Chapter 4 Infrastructure Competition and Open Access Regulation
F IGURE 4.4: Equilibrium prices under NR and MSR.
does not provide a viable wholesale offer to Firm D. Atiyas et al. (2015) and Bourreau
et al. (2011) show that Firm A prefers foreclosure of Firm D if g > g := 26.77. In this case
Firm A sets a foreclosure wholesale price of a NR >
5g+6
.
g2 +7g+6
The left panel of Figure 4.4
depicts ensuing equilibrium retail prices. Note that, due to the foreclosure of Firm D,
the softening effect disappears for g > g and both integrated firms charge identical
retail prices. Furthermore, Firm A engages in a margin squeeze for all g > 3.
Consequently, under MSR the margin squeeze condition is binding iff g > 3. Thus, MSR
effectively prevents foreclosure of Firm D, because Firm A is now required to make a
viable wholesale offer. The resulting equilibrium prices are shown in the right panel of
Figure 4.4.
Figure 4.5a depicts the net effect of MSR on prices, quantities, and profits. If retail goods
are sufficiently differentiated so that foreclosure is not an equilibrium, i.e., g  g, the net
effects are similar to those observed in the previous scenario. The only exception is that,
in this scenario, MSR induces the access provider to reduce its wholesale price relative
to NR. However, similar to the previous scenario, Firm D increases its retail price due
to the stronger strategic complements effect in the retail market. If, instead, retail goods
are less differentiated, i.e., g > g, Firm A’s retail price increases and Firm B’s retail
price decreases under MSR relative to NR. The opposite holds for the integrated firm’s
112
( B ) Firm A’s total, upstream, and downstream profits.
( C ) Consumer surplus, producer surplus, and total surplus.
( A ) Firms’ prices, quantities, and profits.
F IGURE 4.5: MSR effect on market variables in the scenario with simultaneous retail pricing.
4.3 Simultaneous retail pricing
113
Chapter 4 Infrastructure Competition and Open Access Regulation
retail demand. Since MSR prevents Firm D’s foreclosure, its price, demand, and profit
are strictly positive. With respect to Firm A’s profit, the net effect of MSR is ambiguous.
Whereas for g 2 (3, 34.20) increasing wholesale revenue outweighs losses in retail profit
so that Firm A benefits from the regulation, the opposite holds if retail goods are very
close substitutes, i.e., g > 34.20 (see Figure 4.5b).
The welfare analysis of MSR for this scenario is depicted in Figure 4.5c and indicates
similar effects as in the first scenario for g  g. With less differentiated retail goods, i.e.,
g > g, consumers are better off than with rather differentiated goods due to the fore-
closure ban owed to MSR, however, they are still worse off than under NR. In contrast,
producers are better off under MSR due to increased profits of Firm B and Firm D. Remarkably, the increase in producer surplus compensates the loss in consumer surplus,
yielding an increase in total surplus for g > g. Thus, MSR is beneficial from a total
welfare perspective in case of foreclosure—however, this effect emerges from increased
producer surplus, and not from increased consumer surplus.
4.4 Discussion
This chapter scrutinizes margin squeeze regulation in the presence of infrastructure
competition and monopolistic wholesale access based on an analytic game-theoretic
model. More specifically, a setting with two integrated firms of which one offers its
wholesale good whereas the other does not, and one non-integrated firm that depends
on the wholesale good as an input to produce its retail good is considered. In contrast to
a market with a single integrated firm, it is shown that under infrastructure competition
the access provider may engage in a margin squeeze also in the case of an equally
efficient retailer. The central finding is that margin squeeze regulation is detrimental to
consumers, irrespective of the substitutability of retail goods and the timing of firms’
decisions. This result is supported and extended by several insights.
114
4.4 Discussion
First, if the access provider has no incentive to foreclose its downstream rival, margin
squeeze regulation unanimously increases wholesale and retail prices, which leads to a
loss in consumer surplus and ultimately even total surplus. Second, if foreclosure is an
equilibrium, margin squeeze regulation prevents the market exit of the non-integrated
retailer. Although this may lead to a decrease in the wholesale price and an increase
in total surplus, this does not translate into consumers’ benefit, but only into higher
firms’ profits. Third, margin squeeze regulation benefits all firms individually. The
only exception is that the access provider may be worse off under the regulation if
and only if it wants to foreclose the downstream firm and if the retail goods are close
substitutes. In this case, the access provider makes less profit than its integrated rival,
which is likely to evoke non-price discrimination (cf. Mandy and Sappington, 2007).
These findings bear important policy implications. Note that antitrust investigations of
margin squeezes and corresponding regulation are largely enacted in markets where
multiple integrated firms produce an upstream input good, i.e., in industries with infrastructure competition next to service-based downstream competition. In fact, the
European Commission (2013a) argues that infrastructure competition is a necessary
condition for margin squeeze regulation to replace traditional access price regulation.
This rationale is based on the conjecture that retail prices are already constrained by
competition under these circumstances and thus the access provider is compelled to
lower its wholesale price to comply with the requirements of the regulation. The presented findings indicate that this reasoning is flawed if firms compete in prices that are
strategic complements, which is likely to apply to network industries such as telecommunications, but also to other industry contexts. In this case, competitors raise their
retail prices in anticipation of the access provider’s constrained pricing ability. Furthermore, a positive total welfare effect of margin squeeze regulation only occurs if the
non-integrated retailer is foreclosed under no regulation. However, even then the integrated firm which does not provide access has a competitive advantage over the access
provider. Thus, margin squeeze regulation reduces the incentives to provide access
which is contrary to its original rationale as an Open Access rule. Moreover, European
authorities currently do not distinguish between (non-)foreclosure scenarios in margin
115
Chapter 4 Infrastructure Competition and Open Access Regulation
squeeze investigations. Even if they did, the presented results raise skepticism with
regard to the alleged goal of margin squeeze regulation, i.e., to establish a level playing
field for competition, as well as to the ultimate objective of regulatory intervention, i.e.,
to protect consumers.
The analyses conducted in this chapter are limited to price competition in the retail
market. With competition in quantities decisions constitute strategic substitutes, which
may affect the mechanics of margin squeeze regulation. Yet, in Appendix A.2, a model
with differentiated quantity competition à la Singh and Vives (1984) is considered,
which yields similar welfare implications with respect to margin squeeze regulation.
Furthermore, the investigations here abstract from cost asymmetries between integrated firms and non-integrated retailers as well as different application contexts of
margin squeeze regulation in ex post antitrust on the one hand and ex ante sectorspecific regulation on the other hand. This latter issue has been discussed thoroughly
from a competition law perspective (Geradin and O’Donoghue, 2005; Heimler, 2010).
Whereas authorities’ objective—and therefore their assessment of market outcomes—
may differ in these application contexts, the presented economic effects arise irrespective of an ex post or ex ante application of the margin squeeze rule. With regard to cost
asymmetries, Gaudin and Saavedra (2014) summarize the debate on whether a margin
squeeze test should be based on an Equally Efficient Operator or a Reasonably Efficient Operator standard. As the results in this chapter apply to the stricter standard of an equally
efficient retailer, the consideration of a less efficient retailer would further worsen market outcomes from the view of consumers in light of yet higher retail prices. Taking
into account additional complexity (e.g., due to non-linear pricing or bundling), which
in practice is likely to increase the number of false-positive findings of margin squeeze
conduct (Ergas et al., 2010), augments the issues that already arise in the simplified
setting of this study.
With respect to future work, the ineffectiveness of margin squeeze regulation to increase consumer surplus calls for alternative regulatory approaches in cases when there
is infrastructure-based and service-based competition at the same time. It has been ob-
116
4.4 Discussion
served that it is not desirable to rely on (symmetric) access price regulation in this case
(cf. Bacache et al., 2014), because the inherent trade-off between static and dynamic efficiency is likely to stifle investments in infrastructure (see the survey with respect to
OA in Chapter 2). Instead, wholesale competition between the integrated firms may
be seen as a potential alternative to safeguard low input prices for non-integrated competitors and to establish a level playing field in the retail market, although theoretical
research points to cases where competition at the upstream level may fail to improve
market outcomes relative to the monopoly case (Bourreau et al., 2011). Thus, further
research devoted to the design and empirical evaluation of new regulatory institutions
in an environment of competing infrastructures seem to be highly relevant from an
academic and policy perspective. Therefore, the next chapter is dedicated to an experimental analysis of Open Access mechanisms, which encompasses both the test of an
established regulatory institution, in particular margin squeeze regulation, under different market conditions as well as the evaluation of a more behaviorally oriented and
new regulatory instrument, in particular price commitment.
117
Chapter 5
Wholesale Competition and Open Access
Regulation
R
EGULATION of wholesale access to an upstream bottleneck resource that represents an essential input for non-integrated firms to compete in the retail market
downstream stimulates considerable economic research. The anti-competitive effects
that possibly arise in such a scenario as well as accompanying regulatory remedies
(Armstrong et al., 1996; Armstrong and Vickers, 1998) are widely studied in the literature for the case of a single access provider. Moreover, firms’ strategic incentives to
vertically integrate across retail and wholesale markets and the effect of such conduct
on competition (see Lafontaine and Slade, 2007, for an overview) have been thoroughly
investigated, in particular with respect to the softening of retail competition (Chen,
2001; Gans, 2007) and foreclosure of non-integrated retailers (Hart and Tirole, 1990; Ordover et al., 1990; Choi and Yi, 2000; Rey and Tirole, 2007b) or upstream rivals (Chen
and Riordan, 2007). Economic laboratory experiments have complemented the theoretical literature, most notably concerning the issue of vertical foreclosure (Martin et al.,
2001; Normann, 2011). However, a set of new issues arises when there is more than one
vertically integrated access provider such that competition at the wholesale level may
emerge in addition to retail competition. Especially in the likely case of a highly conThis chapter is based on joint work with Niklas Horstmann and Jan Krämer (Horstmann et al., 2016c).
119
Chapter 5 Wholesale Competition and Open Access Regulation
centrated (duopoly) wholesale market the question arises whether access regulation is
(still) warranted. Evidently, the answer to this question will have direct ramifications
on how regulators and competition authorities should deal with this kind of market
structure, but also on whether authorities should promote the entry of a second integrated access provider in markets in which the essential input is currently supplied
monopolistically.
This chapter scrutinizes the effect of wholesale competition on market performance in
terms of market prices, firms’ profits and consumer surplus by explicitly taking into account the emergence of tacit collusion that may arise in this scenario. Based on a framework of two integrated firms and one non-integrated retailer (Bourreau et al., 2011), an
economic laboratory experiment is designed that allows to empirically observe market
performance under various modes of wholesale competition while keeping all other
factors fixed. Moreover, a continuous time framework is employed, which allows firms
to change and observe wholesale and retail prices at any time, and thus, endogenizes
the timing of the price setting (see Section 3.3).
More specifically, the study considers three different market scenarios: First, the case
where only one of the integrated firms provides wholesale access (access monopoly)
is examined. This constitutes the benchmark case, which is extensively studied in the
literature. Second, standard homogeneous Bertrand competition between the two integrated firms at the wholesale level is considered. In this setting, firms can adjust their
wholesale prices at any time and the firm that offers the lower price serves the entire
wholesale market. Third, as Bertrand competition is known to be susceptible to tacit
collusion (Potters and Suetens, 2013), a variant of Bertrand competition at the wholesale level is additionally considered in which integrated firms are obliged to maintain
their wholesale price for a fixed period of time (i.e., a price commitment), everything
else being equal to the second case. Due to this price commitment, the firm that decides
on the lower price is granted a wholesale monopoly position for some time. Thus, this
latter treatment induces an element of competition for the market (Geroski, 2003), which
120
is conjectured to hinder tacit collusion and to intensify competition at the wholesale
level.
All three modes of wholesale competition are examined both under a no regulation
regime, where firms are free to set wholesale and retail prices, and under a margin
squeeze regulation regime, in which an integrated firm’s wholesale price may not exceed its retail price. As described in the previous chapter, margin squeeze regulation
has recently gained attention in the debate on Open Access policies and is perceived
as a viable alternative to price regulation, e.g., by the European Commission (2013a),
particularly when there is more than one wholesale access provider.
The analyzed issues arise most prominently in network industries such as the telecommunications (see Chapter 2) and energy (Boots et al., 2004) industries, in which the
bottleneck arises naturally through subadditivity of the cost structure. Yet, access to an
upstream resource is also of concern in other contexts such as the licensing of intellectual property (Dewatripont and Legros, 2013), where the bottleneck is constituted artificially. Although not confined to this context, the relevance of the considered market
scenario of competing access providers can be exemplified by the telecommunications
industry. Due to technological progress and consolidation both the fixed and the mobile
industries are characterized by few vertically integrated firms, as well as several nonintegrated retailers that rely on access to an upstream resource. On the one hand, with
respect to fixed networks, technological progress led to the roll out of new fiber-optic
networks as well as the evolution of broadband cable networks, which both created
new vertically integrated firms that compete most notably in densely populated urban
areas with the traditional telecommunications incumbent. On the other hand, mobile
telecommunications markets recently experienced a wave of mergers and acquisitions
that reduced the number of independent operators maintaining a distinct cellular infrastructure, thus increasing market concentration at the wholesale level.
Despite its practical relevance, the explicit analysis of simultaneous wholesale and retail competition in the presence of both vertically integrated and non-integrated firms
received little attention in the literature. The extant theoretical analyses, which are re-
121
Chapter 5 Wholesale Competition and Open Access Regulation
viewed in detail below, suggest that wholesale competition is likely to improve and
not deteriorate market performance compared to the case of a wholesale monopoly,
although monopoly-like equilibria may exist. Thereby, the theoretical models generally rest on the assumption of effective competition at the wholesale level, in particular
by assuming homogeneous Bertrand competition between duopolistic access providers
(see, e.g., Bourreau et al., 2011). However, empirical results from both laboratory (Engel, 2007; Potters and Suetens, 2013) and field studies (see, e.g., Parker and Röller, 1997,
in the context of telecommunications markets) suggest that duopoly markets are prone
to high levels of tacit collusion, which may give rise to market outcomes that differ from
those identified in the theoretical literature (see also Chapter 6).
The results of this study indicate that, over and beyond the findings of the theoretical
literature, wholesale competition may in fact lead to a worse market outcome for consumers than a wholesale monopoly. For the case of standard Bertrand competition at
the wholesale level both wholesale as well as retail market prices are above the level
that is observed when there is only a single access provider. Drawing on the literature on upstream collusion (Nocke and White, 2007; Normann, 2009) it is shown that
incentives for tacit collusion are actually higher under wholesale competition if an infinitely repeated game context is considered. Thus, even in the presence of wholesale
competition regulators should closely monitor the performance of such vertically related markets. However, the results also demonstrate that wholesale competition may
be intensified by a simple price commitment rule, which in turn restores the theoretical
prediction to the extent that access prices are lower than under a wholesale monopoly.
Nevertheless, even in this case, wholesale access prices remain well above the predicted
Nash equilibrium. Furthermore, in the context of the Open Access debate, there is no
evidence that a margin squeeze regulation reduces retail market prices compared to a
no regulation regime, which is in line with the findings for a wholesale monopoly in
Chapter 4. Although margin squeeze regulation may benefit the retailer, it tends to
increase retail prices and thus reduce consumers’ surplus.
122
5.1 Related literature
The remainder of this article is structured as follows. Section 5.1 surveys the related
literature on wholesale competition as well as recent studies which have dealt with the
margin squeeze rule. In Section 5.2, the experimental design is described and hypotheses are derived from the theoretical predictions for four timing variants of the basic
model. Section 5.3 presents the experimental results. In Section 5.4, results are examined with respect to the hypotheses and incentives for tacit collusion in a repeated game
context are discussed. Finally, Section 5.5 identifies limitations and points out possible
extensions.
5.1 Related literature
5.1.1 Wholesale monopoly and competition
Before reviewing the literature on wholesale competition, it is worth noting some of
the effects that arise already in the presence of a monopolistic access provider. Even
in the absence of regulation a vertically integrated firm may be willing to supply the
wholesale market on a voluntary basis if the additional revenues generated at the upstream level exceed the business stealing effect of the retailer in the downstream market
(Farrell and Weiser, 2003; Höffler and Schmidt, 2008). More generally, if downstream
organizations exhibit efficiency advantages or if retail goods are sufficiently qualitydifferentiated (e.g., due to brand reputation or additional sales channels as illustrated
by Banerjee and Dippon, 2009), the provision of wholesale services will allow the integrated firm to generate additional revenues. In this case, the access provider benefits
from a demand expansion effect relative to a situation where the integrated firm is the
single seller of its goods in the retail market (Boudreau, 2010).
In the presence of wholesale competition the incentives to provide access on a voluntary basis are likely to be increased compared to a wholesale monopoly, because
the integrated firms may now find themselves in a prisoner’s dilemma with respect to
the provision of the wholesale good (Brito and Pereira, 2010). Studies that investigate
123
Chapter 5 Wholesale Competition and Open Access Regulation
these incentives have examined the conditions under which retailers are supplied in
equilibrium and whether resale actually improves downstream market performance,
particularly in terms of market prices. The majority of these studies considers price
competition with horizontally differentiated retail products (Ordover and Shaffer, 2007;
Brito and Pereira, 2010; Höffler and Schmidt, 2008; Bourreau et al., 2011; Atiyas et al.,
2015) where competition is either spatial (Hotelling, 1929; Salop, 1979) or non-spatial
(Shubik and Levitan, 1980). The remainder assumes quantity competition in the retail
market (Dewenter and Haucap, 2006; Kalmus and Wiethaus, 2010). Although the precise nature of the supply and non-supply equilibria as well as the retail equilibria that
emerge under wholesale competition depend on the specific modeling assumptions, all
theoretical investigations agree that wholesale competition neither leads to more foreclosure of the retailer nor increases wholesale or retail market prices in comparison to
a wholesale monopoly.
More specifically, under wholesale competition Ordover and Shaffer (2007) as well as
Brito and Pereira (2010) find that integrated firms provide the retailer with the retail
good at marginal cost if products are sufficiently differentiated, although they would
be individually better off without entry as retail prices and profits decrease. On the
contrary, retailers are generally not supplied if retail products are close substitutes and
none of the integrated firms has an incentive to make a profitable wholesale offer in the
first place. Furthermore, Ordover and Shaffer show that the supply equilibrium disappears if input goods are differentiated, or if the retailer chooses its quality endogenously
and cannot commit ex ante to its product positioning.
Moreover, the analyses by Brito and Pereira (2010) and Höffler and Schmidt (2008)
reveal that if competition is spatial and the degree of quality differentiation is intermediate, one integrated firm may provide access while the other integrated firm
makes an unprofitable offer. This finding of a partial foreclosure equilibrium is further
generalized—including the case of non-spatial competition—by Bourreau et al. (2011)
based on the characterization of the softening effect: A vertically integrated wholesale
provider chooses its retail price with regard to its opportunity costs in the wholesale
124
5.1 Related literature
market (DeGraba, 2003) and thus will be less aggressive in the retail market than its
vertically integrated rival who does not provide wholesale access. In other words,
the consideration of opportunity costs weakens competition in the retail market and
may at the same time make it less attractive to compete for wholesale revenues. In
consequence, the monopoly outcome may be restored, because the integrated rival of
the access provider benefits from higher retail profits and thus prefers to exit the upstream market. Note, however, that the equilibrium hinges on the assumptions that
retail goods are close substitutes and that at least one firm supplies the retail firm, e.g.,
due to a retailer’s efficiency advantage (Bourreau et al., 2011) or due to regulatory coercion (Bourreau et al., 2015). Otherwise, marginal cost pricing in the upstream market constitutes the unique equilibrium under wholesale competition in the non-spatial
model if goods are sufficiently differentiated (Höffler and Schmidt, 2008). Moreover,
Atiyas et al. (2015) show that unobservable, more complex wholesale contracts may
stimulate voluntary access and wholesale competition, thus making foreclosure of the
retailer less likely.
Höffler and Schmidt (2008) investigate the effects of resale on consumer welfare, which
may be increased either by a decline in retail prices and/or an increase in variety. Under
the assumption that the retailer will be supplied by one of the integrated firms, i.e.,
there is no foreclosure, it is shown that resale may actually increase the market price
if quality differentiation is sufficiently high. In the case of non-spatial competition the
price increase is always compensated by an increase in variety with respect to consumer
welfare. In the spatial model however, consumers may be worse off as the price effect
dominates. Then again, if wholesale competition for retailers is considered in the nonspatial model, wholesale prices are found to equal marginal cost and, in consequence,
retail prices are lower than compared to a situation without resale.
Whereas the reported analyses of wholesale competition focus exclusively on the oneshot interaction between firms, the literature on upstream collusion examines incentives for coordinated firm behavior in an infinitely repeated game setting. Nocke and
White (2007) compare critical discount factors that are necessary to sustain collusion
125
Chapter 5 Wholesale Competition and Open Access Regulation
by the means of grim trigger strategies and find that vertical integration facilitates tacit
collusion among upstream firms relative to a vertically separated industry structure.
Normann (2009) replicates the finding that vertical integration facilitates upstream collusion for the case of linear input charges and a sequential setting of wholesale and retail
prices, whereas Nocke and White model wholesale contracts as two-part tariffs and assume simultaneous price setting.
This study contributes to the literature on wholesale competition by showing empirically that wholesale prices may be above the monopoly level even if theory predicts
wholesale supply at marginal costs as the unique equilibrium. It is further shown that
tacit collusion at the wholesale level may effectively be reduced by a price commitment
rule that fosters the integrated firms’ competition for the market. Moreover, the experimental framework allows for a systematic comparison of retail market performance
in terms of wholesale and retail prices, firms’ profits and consumer surplus under the
different modes of wholesale competition. Finally, the experimental design in continuous time endogenizes the timing of firms’ price setting, and thus reconciles different
timings proposed in the theoretical literature.
5.1.2 Margin squeeze regulation
In the presence of a duplicate infrastructure the traditional economic rationale for ex
ante price regulation is no longer applicable as the bottleneck does not represent a
single essential facility anymore (Renda, 2010). In consequence, regulators and competition agencies may be concerned with identifying suitable alternatives and regulatory rules that still ensure Open Access for downstream competitors, but give integrated
firms more freedom in setting their wholesale prices (see Chapter 2). As described in
Chapter 4, the margin squeeze rule represents a potential surrogate for price regulation
that is already applied in various forms and different contexts. Next to its application
in (European) competition law, the basic mechanism, which is designed to ensure a
viable wholesale-retail margin for a downstream retailer, is also implemented by re-
126
5.1 Related literature
tail minus X regulation (Gonçalves, 2007) and the efficient component-pricing rule (Baumol
et al., 1997). Ever since the landmark decision Deutsche Telekom1 in 2003, the application
of the margin squeeze rule as an antitrust instrument is controversially debated within
the economic and the legal literature (Briglauer et al., 2011; Carlton, 2008; Geradin and
O’Donoghue, 2005). While the European Commission has repeatedly convicted firms
based on a margin squeeze accusation2 and has been confirmed by European courts3 ,
the US Supreme Court has dismissed allegations based on the margin squeeze rationale
in comparable cases (Trinko and linkLine).
The rationale for margin squeeze regulation is that protecting competitors in the context of monopolistic bottlenecks or concentrated input markets will ultimately benefit
consumers. Particularly in competition policy the latter goal is emphasized and held in
high regard, e.g., the European Commission (2009a, p.7) clarifies that “what really matters is protecting an effective competitive process and not simply protecting competitors”. Sector-specific regulation may widen the scope of application, as is illustrated
by the debate about the relevant efficiency standard for the margin squeeze conduct
(Geradin and O’Donoghue, 2005), but fundamentally still aims at the protection of consumers, where competition itself is a means to an end (Vogelsang, 2013).
In this vein, Jullien et al. (2014) provide an overview of the economic theories of harm
that may qualify a margin squeeze as an abuse of market power and could provide
the basis for a stand-alone antitrust doctrine. Petulowa and Saavedra (2014) qualify
the circumstances under which a margin squeeze can occur in the case of differentiated
goods and state that a margin squeeze is rather the result of competition and not of an
exploitative abuse. Jullien et al. conclude that the effects of a margin squeeze rule are
ambiguous as wholesale prices may decrease, but retail prices may also rise, due to a
price umbrella effect. With regard to retail minus X regulation, Höffler and Schmidt (2008)
criticize that its application may lead to consumer welfare losses and higher prices.
1 Commission
Decision 2003/707/EC.
the Commission Decision of 4 July 2007 (Case COMP/38.784 – Wanadoo España vs. Telefónica).
3 See the cases Deutsche Telekom (T-271/03, C-280/08), Telefónica (T-336/07, T-398/07 C-295/12) and TeliaSonera (C-52/09).
2 See
127
Chapter 5 Wholesale Competition and Open Access Regulation
In the past, the margin squeeze rule has mostly been investigated in the case of a single access provider, as indispensability has initially constituted a central criterion in
its application as an antitrust instrument. More recently, as illustrated by the ex ante
economic replicability test in the European Commission’s 2013a recommendation on consistent non-discrimination, the margin squeeze test may also be applied to an environment with competing infrastructures (Jaunaux and Lebourges, 2015). Although this
rule is already applied in practice, little research has been conducted with regard to actual consequences in the particular application context of infrastructure and wholesale
competition. This study complements the theoretical analysis of margin squeeze regulation in the context of infrastructure competition with an experimental evaluation and
further extends the examination to the context of wholesale competition.
The experimental evidence obtained in this study suggests that the margin squeeze rule
is likely to be ineffective in lowering retail prices irrespective of the mode of wholesale
competition. Although margin squeeze regulation may benefit the retailer in some circumstances, it tends to increase retail prices and thus reduces consumer surplus.
5.2 Experimental framework
5.2.1 Theoretical model
The underlying experimental framework explicitly addresses the presented issues of
wholesale competition and Open Access by incorporating a market design that allows
for competition at the wholesale and retail level. The general experimental design
is based on the model of upstream competition analyzed by Bourreau et al. (2011)—
illustrated in Figure 5.1— in which two integrated firms (Firm A & Firm B) are able
to supply the wholesale good, while a third firm (Firm D) operates only in the downstream market. In order to supply the retail good, the downstream retailer is required
to purchase the wholesale good at the upstream market from one of the two integrated
128
5.2 Experimental framework
F IGURE 5.1: Conceptual model of wholesale competition with two integrated firms and a nonintegrated retailer (based on Bourreau et al., 2011, p. 683).
Firm B
Firm A
Wholesale
market
Wholesale
good
Retail
market
aB
aA
Wholesale
good
Firm D
Retail
good
pA
Retail
good
pD
Retail
good
pB
Consumers
firms. The wholesale prices of Firm A and Firm B are denoted a A and a B , respectively.
In the retail market, all firms choose their respective retail prices pk , k 2 { A, B, D }.
It is assumed that Firm D chooses the wholesale product with the lowest price and
does not split its demand.4 Thus, the integrated firms compete à la Bertrand with homogeneous goods. For each quantity that the downstream retailer supplies to consumers in the retail market it must buy an identical quantity of the wholesale good.
In the downstream market, firms compete likewise in prices, but goods are differentiated. In line with previous theoretical studies on wholesale competition, competition in horizontally differentiated goods based on Shubik and Levitan (1980)5 is assumed, where the retail demand of each firm k in the case of n = 3 active firms is given
by qk = 13 (1
pk
g( pk
Â3i=1 pi
3 ))
and the differentiation parameter g defines the de-
gree of substitution between firms’ retail goods. Across all treatments, g = 30, which
corresponds to a diversion ratio of 10/21 for each pair-wise relationship between firms
(Shapiro, 1996).
4 Note
that, as the stage game is played repeatedly in the experiment, the following tie breaking rule is
used: If Firm A and Firm B offer the same wholesale price, Firm D chooses to purchase access from the
firm that has previously offered the lower price. If both integrated firms offer an identical wholesale
price in the first period, the access provider is chosen randomly.
5 The model assumes that consumers explicitly value variety. With regard to the derivation of the demand
structure for varying numbers of active firms in the market see Höffler (2008).
129
Chapter 5 Wholesale Competition and Open Access Regulation
Throughout all experimental sessions, Firm D is modeled to mimic the behavior of a
competitive fringe in the retail market that reacts to the price setting by the integrated
firms, Firm A and Firm B. It is therefore assumed that Firm D always chooses the bestresponse retail price, i.e., the price that maximizes its profit given the wholesale and
retail prices set by the integrated firms.
In the experiment, prices are scaled as follows: Values obtained by the Shubik and Levitan (1980) model are multiplied by 100/0.15. and firms can set their prices to any integer
in the range of zero to one hundred. In terms of the original Shubik and Levitan (1980)
values, this corresponds to the price interval [0; 0.15]. As a consequence, the joint profit
among integrated firms’ is maximized when integrated firms choose maximum prices
in both the wholesale market (amax = 100) and the retail market (pmax = 100). Therefore, the JPM outcome is identical across treatments. Moreover, the scaling allows for
a more granular representation of the relevant price interval between the theoretically
predicted competitive and collusive prices as well as the monopoly price in the case of
a single access provider.
In contrast to the theoretical literature on wholesale competition, which usually prescribes a specific temporal sequence of actions, timing of price decisions is endogenized
in the experiment by means of a continuous time framework. Endogenous timing of
price setting has two aspects: First, no assumption is made on the sequence of upstream
and downstream decisions. While it is frequently assumed that wholesale prices are set
prior to retail prices (Bourreau et al., 2011), prices may also be chosen simultaneously
(Nocke and White, 2007). Second, price setting of the integrated firms at a specific market level is equally unconstrained, i.e., these firms decide not only about the magnitude
of a price, but also about timing when to change it. Therefore, the experimental design
includes various time settings that are captured by the theoretical literature, but at the
same time allows for a more general approach as it also incorporates additional settings
that may arise endogenously. In consequence of the endogenous timing induced by the
continuous time framework, multiple theoretical predictions may apply, depending on
the specific temporal sequence of firms’ actions. In order to provide a robust theoreti-
130
5.2 Experimental framework
cal prediction, consider the following four alternative timing models, which are variants
of either a sequential-move or a simultaneous-move game proposed in the theoretical
literature:
(1) Two-stage game as suggested by Bourreau et al. (2011)6 : First, integrated firms
set their wholesale prices simultaneously and the downstream retailer chooses its
access provider. Second, all firms decide simultaneously on their retail prices.
(2) Three-stage Stackelberg game: Same as (1) with the exception that the downstream retailer chooses its retail price in a third stage, i.e., after the integrated
firms have chosen their respective retail prices.
(3) Simultaneous-move game as assumed by Nocke and White (2007): All firms set
all of their prices, both wholesale and retail, simultaneously.
(4) Two-stage Stackelberg game: Same as (3) with the exception that the downstream
retailer chooses its retail price in a second stage, i.e., after the integrated firms
have chosen their prices.
Table 5.1 denotes theoretical equilibrium predictions for four market scenarios and depicts resulting ordinal differences of wholesale and retail prices for all four timing models. Note that the hypotheses regarding the direction of a price difference hold equally
for wholesale and retail prices of each individual firm. Although the timing models
vary with regard to the specific numerical predictions for equilibrium prices in the investigated scenarios, the direction of price effects between scenarios align—with one
exception that is discussed below.7
6 Note that, in contrast to Bourreau et al. (2011), there is no assumption made that the downstream retailer
will always be supplied by at least one integrated firm and therefore consider cases where complete
foreclosure may arise as an equilibrium (see Atiyas et al., 2015, for a detailed analysis in context of
the Shubik and Levitan model). Integrated firms may choose to set wholesale prices in excess of their
own retail prices and consequently foreclose the retailer from the downstream market, which implies
q D = 0.
7 See Appendix A.3 for the complete analysis and a comprehensive comparison of all models.
131
Chapter 5 Wholesale Competition and Open Access Regulation
TABLE 5.1: Predicted wholesale and retail price differences between market scenarios.
Wholesale monopoly
No regulation
Margin squeeze regulation
Monopoly outcome
or
Foreclosure
Constrained
monopoly
outcome

>
Wholesale competition
Competitive outcome
or
Foreclosure
Competitive outcome
In order to allow for a benchmark for the evaluation of wholesale competition, consider
first the market outcome under a wholesale monopoly. In this scenario only Firm A offers a wholesale price and may provide the wholesale good to the retailer Firm D. By
contrast, Firm B relies on its vertically integrated structure to produce its own wholesale good, but does not offer access to its wholesale resource. In the absence of regulation, given g = 30, Firm A is expected to set the wholesale price either at the wholesale
monopoly level a A = am or such that the retailer is foreclosed from the downstream
market. As shown in Appendix A.3, the latter outcome arises if the retailer’s price reaction is not explicitly anticipated by the monopolistic wholesale provider, i.e., in Timing
Models (1) and (3). The introduction of margin squeeze regulation changes the equilibrium outcome in the wholesale monopoly scenario only if equilibrium prices of Firm
A under no regulation violate the margin squeeze condition a A  p A . Whereas margin
squeeze regulation is then expected to decrease prices in comparison to the foreclosure
outcome, it instead increases wholesale and retail prices for all firms according to the
theoretical prediction in Timing Model (2), where foreclosure does not constitute an
equilibrium under no regulation. In sum, theoretical predictions on whether the implementation of margin squeeze regulation decreases prices in a wholesale monopoly are
ambiguous.
In the wholesale competition scenario, it is straightforward that symmetric marginal
cost pricing, i.e., a A = a B = 0, is a Nash equilibrium, as is shown by Bourreau et al.
(2011). The corresponding equilibrium retail prices are thus symmetric for all three
firms. In Timing Models (2) and (4) this equilibrium is unique, because integrated firms
132
5.2 Experimental framework
anticipate that Firm D as a follower can only act as a price taker and therefore find it
always profitable to make a viable wholesale offer.8 In contrast, in Timing Models (1)
and (3), there exists a second foreclosure equilibrium in which both integrated firms
decide not to offer a viable wholesale price to the retailer, i.e., the retailer does not supply any retail consumers (Atiyas et al., 2015). Introducing margin squeeze regulation in
the case of wholesale competition renders foreclosure impossible, thus, the competitive
equilibrium remains as the unique predicted outcome in all presented timing models.
In conclusion and in line with previous theoretical analyses, prices under wholesale
competition are likely to be below prices in a wholesale monopoly and never exceed
them across all model variants.
5.2.2 Design
The experimental design is based on a continuous time framework in which participants can observe competitors’ price changes immediately and market variables are
updated in real time. Similar designs have recently been used in experimental economics, e.g., in the context of a prisoner’s dilemma game (Bigoni et al., 2015; Friedman
and Oprea, 2012) as well as in a Hotelling setting (Kephart and Friedman, 2015). Next
to its property to endogenize the timing of the game and thereby to reconcile different
timings proposed in the theoretical literature, the continuous time framework is chosen
for the following reasons: First, continuous time is conjectured to promote the emergence of a theoretical prediction in complex market settings (Kephart and Friedman,
2015; Kephart and Rose, 2015). Second, under both Cournot as well as Bertrand competition, Section 3.3 systematically compares the extent of tacit collusion that emerges
under continuous time and discrete time and finds lower levels of tacit collusion in continuous time for both competition models. Therefore, the continuous time framework
offers a more conservative experimental test of the emergence of tacit collusion than
a discrete time framework. Third, through the continuous feedback loop subjects can
8 Hence, modelling the retailer as a follower in a Stackelberg retail setting may be viewed as an alternative
implementation of the a priori assumption made by Bourreau et al. (2011) which guarantees that the
integrated firms have no incentive to foreclose the retailer.
133
Chapter 5 Wholesale Competition and Open Access Regulation
directly assess the interdependency between prices in the wholesale and retail market,
which aids them in evaluating the impact of their decisions on their individual performance and on aggregate market outcomes.
The experiment is computerized with the Java-based experimental software Brownie
(Müller et al., 2014). The course of the experiment is divided in two phases: the trial
phase and the game phase. During the trial phase subjects are able to test various price
configurations for all firms in the particular market cohort and to observe the resulting
payoffs, while these actions do not impact the subjects’ earnings and are not visible
to other participants, i.e., the subjects do not interact with each other during the trial
phase. The game phase, which starts after all subjects confirm their initial prices in the
trial phase, lasts for exactly 30 minutes. All decisions in the game phase directly impact
the monetary payoff of the subjects. Earnings are the cumulative profits over the time
horizon of the experiment. Current profits and cumulative earnings are displayed to
subjects over the entire game phase.
As motivated above, the integrated firms, Firm A and Firm B, are represented by human subjects while the downstream retailer, Firm D, is represented by an automated
software agent. The agent is programmed to constantly choose its profit-maximizing
price given the wholesale and retail prices set by the integrated firms. Thereby, the software agent reacts immediately to any price change made by one of the other firms. In
this setup, the experiment covers the following three modes of wholesale competition
and two regulatory Open Access regimes in a full-factorial manner, thus ensuing six
treatments (see Table 5.2):
Wholesale Monopoly (WM): Only Firm A sets a wholesale price and can change it at any
time. Firm B does not participate in the wholesale market.
Wholesale Competition (WC): Firm A and Firm B set and can change wholesale prices at
any time.
134
5.2 Experimental framework
TABLE 5.2: Full-factorial experimental design with six treatments.
Treatments
No Regulation
Margin Squeeze Regulation
Wholesale Monopoly
WM-NR
(N = 12)
WM-MSR
(N = 11)
Wholesale Competition
WC-NR
(N = 9)
WC-MSR
(N = 10)
WCPC-NR
(N =10)
WCPC-MSR
(N = 12)
Wholesale Competition with
Price Commitment
Independent observations at the market cohort level in parentheses.
Wholesale Competition with Price Commitment (WCPC)) Firm A and Firm B set wholesale
prices, however, each firm’s wholesale price is fixed for an embargo period of 30
seconds after it is changed, everything else being equal to WC.
No Regulation (NR): Firms set wholesale and retail prices freely.
Margin Squeeze Regulation (MSR): Firm A and Firm B may set neither their wholesale
price above their own retail price nor their retail price below their own wholesale
price. If firms set wholesale (retail) prices that violate these conditions, the experimental software displays a warning and sets the price to the allowed maximum
(minimum), which is the current own retail (wholesale) price.
5.2.3 Procedures
The experimental sessions were conducted with students of the Department of Economics and Management at the Karlsruhe Institute of Technology, Karlsruhe, Germany,
who were recruited via the ORSEE platform (Greiner, 2015). Overall, 128 subjects participated in the study and each participant played only one of the treatments (betweensubject design). The average experimental session lasted 70 minutes. On average, subjects earned a performance-based payment of 16.80 Euro in addition to a base fee of 5
Euro. Participants were randomly assigned to groups of two and interacted with the
same firm for the entire time horizon of the experiment (fixed partner matching). Consequently, there are 64 independent observations at the market cohort level as denoted
by Table 5.2. The current market data is recorded every 500 ms, thus, there are 3,600
135
Chapter 5 Wholesale Competition and Open Access Regulation
data tuples per market cohort that include wholesale and retail prices as well as the
corresponding quantities and profits.
While the main analyses and results will focus on the student sample, a complementary validation study was additionally conducted for the WCPC-NR treatment with 16
professional experts in an effort to address external validity concerns. The experts were
recruited from the regulatory department of a major German telecommunications operator, where they deal with issues of access regulation on a daily basis. The study was
executed under identical conditions as in student experiments with three exceptions.
First, the duration of the game phase was shortened to ten minutes. Second, the payment scheme was changed to a lottery system, where participants could win one of
three vouchers with a monetary value of 30 Euro each. The number of lottery tickets
that participants received were dependent on their payoff in the experiment. By this
means, monotonicity was ensured with regard to the relationship between individual
performance and payoffs. Third, each participant played a second WCPC-NR treatment
with a more differentiated retail market (g = 50). The sequence of the two treatments
was randomized across three experimental sessions.
All experimental sessions with students as well as experts were conducted with the
same experimental software and hardware in order to ensure consistency, particularly
with regard to the graphical user interface. Each session was run according to the following protocol. Upon entering the laboratory, subjects are randomly assigned to a seat,
from which they can neither see nor speak to any other participant of the experiment.
Subsequently, the experimental instructions9 are handed to the participants in print
and read aloud from a recording. Paragraphs that are identical across treatments are
recorded once and the recording is used in all treatments. Prior to the beginning of the
experiment, each subject has to complete a computerized comprehension test that includes a set of questions regarding the experimental instructions and the experimental
procedure. Participants are allowed to proceed to the next question only after entering
the correct answer to the current one. After all subjects successfully complete the test,
9 Exemplary
136
instructions for the WCPC-MSR treatment are provided in Appendix C.2.
5.3 Results
the experiment starts automatically. In addition to this procedure, student participants
wore ear protectors from the beginning of the questionnaire until the end of the game
phase in order to avoid any influence from clicking noises of computer mouses.
5.3 Results
5.3.1 Main study
In the following, market prices, firms’ profits and consumer welfare are evaluated
across treatments for the main study with students. The wholesale market price am
is given by the wholesale price that the entrant faces, i.e., the minimum of both wholesale offers. The retail market price ym is defined as the transaction price, which is the
demand-weighted average of retail prices, i.e., ym = Âk
qk
Q
· pk where Q is the aggre-
gate market demand. Profits are given by the amount of money that participants earn
during the game phase, i.e., the final payoff excluding the fixed base fee. The average
profit of both integrated firms is denoted by p AB and the profit of the downstream retailer by p D . Consumer surplus is computed as the utility of a representative consumer
given the supplied quantities of all three firms subtracted by the transaction price, i.e.,
CS = Âk qk
3
( q2
2(1+ g ) Â k k
+ g3 (Âk qk )2 )
Âk qk pk (see Bouckaert and Kort (2014) for a
detailed derivation). For ease of interpretation consumer surplus is standardized as
f =
CS
CS CSmin
CSmax CSmin
f = 0 deon the interval of eligible prices, i.e., pk 2 [0, 100]. Thus, CS
f = 1 represents the maxinotes the minimum consumer surplus at pk = 100, while CS
mum consumer surplus at pk = 0. For a focus on market outcomes in a stable market
environment and due to the complexity of the experiment start- and endgame effects
are neglected by considering only the market data from recorded ticks 601 to 3,000 with
1 tick = 500 ms, i.e., the first five and last five minutes are dropped for the subsequent
analysis. For the same reasons, the analysis is based on medians as this mitigates the
impact of outliers in comparison to averages and should therefore provide a more conservative analysis (see, e.g., Friedman and Oprea (2012) for an identical approach in a
continuous time experiment). Arguably, regulators and policy makers should be more
137
Chapter 5 Wholesale Competition and Open Access Regulation
TABLE 5.3: Treatment median of median market cohort prices, profits and consumer surplus.
Treatment
f
CS
Markets
N
am
ym
p AB
pD
WM-NR
WM-MSR
WC-NR
WC-MSR
WCPC-NR
WCPC-MSR
12
11
9
10
10
12
28,800
26,400
21,600
24,000
24,000
28,800
73.573
72.572
86.085
83.082
40.540
49.049
65.499
83.124
88.407
92.281
49.560
67.415
16.383
20.750
22.434
22.802
12.243
15.866
0.899
1.028
0.258
2.339
2.491
2.322
0.292
0.153
0.097
0.062
0.461
0.298
Total
64
153,600
72.071
76.159
18.093
1.672
0.213
Medians are based on minutes [5, 25] of the game phase.
interested in the median outcome that can be expected from a single scenario than the
average effect across multiple co-existing scenarios.10
Table 5.3 presents the respective treatment medians of median values at the market
cohort level for wholesale and retail prices, firms’ profits, and consumer surplus together with the number of independent market cohorts and the number of partially
dependent observed time ticks. In addition, Figure 5.2 depicts the period medians of
wholesale and retail market prices across individual market cohorts for each of the six
treatment combinations. For purposes of illustration, every point in the graphs is a
median over 50 subsequent ticks. In order to evaluate treatment effects statistically,
consider the following quantile regression (Koenker and Hallock, 2001):
X jt = b 0 + b Period · t + bWC · WC
+ bWCPC · WCPC
+ b MSR · MSR
+ bWCxMSR · WC · MSR
+ bWCPCxMSR · WCPC · MSR + e jt ,
where X jt denotes the respective market variable X in market cohort j and period t.
Treatment WM-NR is adopted as the baseline.11 WC, WCPC and MSR are dummy vari10 Nevertheless,
the reported results are similar if the analysis is based on means rather than on medians
(see Appendix B.2).
11 Pairwise comparisons between all treatments by means of quantile regressions are reported in Appendix B.2.
138
5.3 Results
F IGURE 5.2: Median wholesale (dashed) and retail (solid) market prices for each of the six treatments.
300 600 900 1200 1500 1800
0
300 600 900 1200 1500 1800
300 600 900 1200 1500 1800
300 600 900 1200 1500 1800
WC-MSR
0
300 600 900 1200 1500 1800
WCPC-MSR
0 20 40 60 80 100
WCPC-NR
0
0
0 20 40 60 80 100
0 20 40 60 80 100
WC-NR
0 20 40 60 80 100
Median market price
0
WM-MSR
0 20 40 60 80 100
0 20 40 60 80 100
WM-NR
0
300 600 900 1200 1500 1800
Time in seconds
139
Chapter 5 Wholesale Competition and Open Access Regulation
TABLE 5.4: Quantile regressions of wholesale market price am , retail market price ym , intef
grated firms’ average profit p AB , retailer’s profit p D and consumer surplus CS.
Baseline: Wholesale Monopoly & No Regulation (WM-NR).
Covariate
Wholesale
Competition (WC)
WC w/ Price
Commitment (WCPC)
(1)
am
(2)
ym
23.879⇤⇤
(10.073)
18.376⇤
(9.738)
4.839⇤⇤⇤
(1.830)
0.813
(0.815)
0.162⇤⇤
(0.080)
19.763⇤⇤
(9.130)
5.459⇤⇤⇤
(1.729)
1.130
(0.861)
0.212⇤⇤⇤
(0.077)
28.154⇤⇤⇤
(8.851)
(3)
p AB
(4)
pD
(5)
f
CS
Margin Squeeze
Regulation (MSR)
9.401
(10.812)
9.721
(9.927)
2.358
(1.827)
0.173
(0.798)
0.081
(0.078)
WC x MSR
24.309⇤
(13.731)
7.156
(12.170)
2.562
(2.403)
1.924⇤⇤
(0.926)
0.060
(0.100)
WCPC x MSR
1.928
(17.188)
7.250
(12.692)
1.759
(2.774)
0.019
(0.930)
0.088
(0.116)
0.001⇤
(< 0.001)
> 0.001
(< 0.001)
Period
0.004
(0.003)
0.004⇤⇤
(0.002)
> 0.001⇤⇤
(< 0.001)
Constant
62.893⇤⇤⇤
(9.764)
62.972⇤⇤⇤
(9.148)
16.251⇤⇤⇤
(1.931)
1.324⇤
(0.769)
0.319⇤⇤⇤
(0.080)
Observations
153,600
153,600
153,600
153,600
153,600
Clustered standard errors (by market cohort) in parentheses.
⇤ p < 0.10, ⇤⇤ p < 0.05, ⇤⇤⇤ p < 0.01
ables indicating the respective mode of wholesale market structure and Open Access
regulation. Interactions WC x MSR and WCPC x MSR delineate the effects of margin
squeeze regulation under a specific mode of wholesale competition. Standard errors
are clustered on the market cohort level to control for intra-cluster correlation over repeated observations from periods in the same market cohort (Parente and Silva, 2016).
The estimates of the respective models for market variables of interest are reported in
Table 5.4 and interpreted in the following.
In the benchmark case WM-NR the wholesale monopolist sets a positive wholesale
price as is indicated by the estimated constant in Model (1), which is similar in magnitude to the retail market price as reported in Model (2). This is in line with the observation that margin squeezes occur frequently, such that the non-integrated firm is
effectively foreclosed, i.e., the wholesale market price is greater than or equal to the re-
140
5.3 Results
tail prices of both integrated firms.12 The median rate of foreclosure at the individual
market cohort level amounts to 49.52% in this scenario. Still, the profit of Firm D is
found to be significantly different from zero, as indicated by the positive constant in
Model (4). In other words, even in the case of an unregulated wholesale monopoly,
the downstream retailer can profitably participate in the retail market. Due to its access
monopoly, Firm A achieves a significantly (p < 0.01) higher median profit (p A = 19.031)
than its integrated competitor (p B = 14.294).13 Note that all of these results are in line
with the theoretical prediction.
R ESULT 5.1. The introduction of wholesale competition reduces neither wholesale prices nor
retail prices. In fact, under homogeneous Bertrand competition at the wholesale level consumers
as well as the downstream retailer are worse off compared to the case of an unregulated wholesale
monopoly.
Surprisingly, relative to an unregulated wholesale monopoly, the introduction of homogeneous Bertrand competition at the wholesale level increases both wholesale and retail
market prices significantly. While under WC-NR the wholesale market price rises by
23.88 pp, consumers face an 18.38 pp higher retail market price in comparison to WMNR. Although it is well-known that Bertrand competition yields supra-competitive
prices, it is notable that under WC-NR prices are set even significantly above price levels
of WM-NR. In consequence, the ability to tacitly collude in the wholesale market allows
the integrated firms to extract higher profits than in the monopoly treatment as indicated in Model (3). While the effect on the retailer’s profit is negative but insignificant,
the median rate of foreclosure is 62.46%, and thus higher than under WM-NR.
R ESULT 5.2. Competition in the wholesale market can be stimulated by introducing competition for the market through a price commitment. Then, wholesale and retail prices are lower
than under a wholesale monopoly, but remain above the theoretical prediction.
12 Note
that the non-integrated firm may still be marginally active in the retail market, because goods are
differentiated.
13 See Appendix B.2 for the corresponding quantile regression.
141
Chapter 5 Wholesale Competition and Open Access Regulation
Remarkably, the collusive effect of wholesale competition is set off by a simple wholesale price commitment for integrated firms. In particular, under WCPC-NR the wholesale market price decreases significantly by 28.15 pp (52.03 pp) relative to WM-NR
(WC-NR), while the transaction price in the retail market is lowered significantly by
19.76 pp (38.14 pp). As a result, consumers’ surplus increases significantly by 21.2
pp compared to WM-NR as indicated by Model (5). In line with declining market
prices, the integrated firms’ profits decrease significantly as well. Despite lower wholesale prices, the margin between wholesale and retail prices remains relatively slim
due to the increasing price competition at the retail level as is depicted by the lower
left panel in Figure 5.2. In consequence, the median rate of foreclosure amounts to
29.10%. The effect on the retailer’s profit is found to be insignificant, although positive
in absolute terms. Evidently, the estimated wholesale access under WCPC-NR price of
62.893
28.154 = 34.739 remains well above the theoretical prediction of am = 0.
R ESULT 5.3. There is no evidence that margin squeeze regulation reduces retail prices, and thus
consumers do not benefit from such a regulation. However, the introduction of a margin squeeze
regulation may reduce wholesale prices, and thus the downstream retailer may be better off.
As reported above, margin squeezes are frequently observed under all market structures at the wholesale level. Since the primary justification for margin squeeze regulation is the prevention of exclusionary and exploitative abuses (Jullien et al., 2014), its
impact on market prices and surplus measures is examined in the following. The regression analyses reported in Table 5.4 reveal that margin squeeze regulation generally
does not have a significant impact on market outcomes, but rather tends to increase
wholesale and retail prices. In fact, the only reduction in wholesale prices evoked by
margin squeeze regulation is found in the case of a particularly collusive wholesale
market as under unregulated wholesale competition. More specific, the wholesale price
under WC-MSR is significantly lower than under WC-NR, which is indicated visually
by the middle panels in Figure 5.2 and supported empirically by the significant interaction effect WC x MSR. Although this effect is paralleled by an increase in the retailer’s
profit, margin squeeze regulation translates neither into significantly lower retail prices
142
5.3 Results
TABLE 5.5: Quantile regressions of market outcomes in case of wholesale competition without
price commitment.
Baseline: Wholesale Competition & No Regulation (WC-NR).
(1)
am
(2)
ym
(3)
p AB
Margin squeeze
regulation (MSR)
14.632⇤
(8.254)
3.909
(3.528)
0.149
(1.034)
Period
0.005
(0.005)
< 0.001
(0.002)
< 0.001
(< 0.001)
Covariate
Constant
Observations
85.979⇤⇤⇤
(15.948)
45,600
(4)
pD
2.051⇤⇤⇤
(0.520)
> 0.001
(< 0.001)
(5)
f
CS
0.036
(0.030)
> 0.001
(< 0.001)
88.125⇤⇤⇤
(5.347)
22.068⇤⇤⇤
(1.638)
0.650
(0.869)
0.105⇤⇤
(0.045)
45,600
45,600
45,600
45,600
Clustered standard errors (by market cohort) in parentheses.
⇤ p < 0.10, ⇤⇤ p < 0.05, ⇤⇤⇤ p < 0.01
nor into significantly higher consumer surplus. Taken together, the empirical results do
not provide any evidence that consumers or the retailer generally benefit from a margin
squeeze regulation.
In an effort to further investigate the impact of the margin squeeze regulation and to
delineate effects on stakeholders under different wholesale competition models, the
following (reduced) quantile regression model is estimated for each of the wholesale
market structures separately to allow for a pairwise comparison:
X jt = b 0 + b Period · t + b MSR · MSR + e jt .
This analysis is of particular interest whenever policymakers are able to prescribe rules
that govern the competition at the wholesale level but may find themselves unable
to change the market structure completely. In these cases the margin squeeze condition may be considered as an ex ante regulatory remedy or as an ex post competition
policy instrument. The effect of margin squeeze regulation is therefore examined under all three considered wholesale market structures. First, in the case of a wholesale
monopoly, margin squeeze regulation has a positive yet insignificant effect for all price
and profit variables, while the corresponding coefficient for consumer surplus is nega-
143
Chapter 5 Wholesale Competition and Open Access Regulation
TABLE 5.6: Quantile regressions of market outcomes in case of wholesale competition with price
commitment.
Baseline: Wholesale Competition with Price Commitment & No Regulation
(WCPC-NR).
(1)
am
(2)
ym
(3)
p AB
(4)
pD
7.983
(10.883)
17.081⇤⇤
(8.071)
4.157⇤
(2.251)
0.188
(0.474)
Period
0.003
(0.004)
0.006⇤
(0.003)
0.001
(0.001)
> 0.001
(< 0.001)
Constant
37.475⇤⇤⇤
(8.131)
39.162⇤⇤⇤
(5.664)
10.374⇤⇤⇤
(1.921)
2.335⇤⇤⇤
(0.609)
0.572⇤⇤⇤
(0.058)
Observations
52,800
52,800
52,800
52,800
52,800
Covariate
Margin squeeze
regulation (MSR)
(5)
f
CS
0.165⇤⇤
(0.080)
> 0.001⇤
(< 0.001)
Clustered standard errors (by market cohort) in parentheses.
⇤ p < 0.10, ⇤⇤ p < 0.05, ⇤⇤⇤ p < 0.01
tive and insignificant (see Appendix B.2). In line with the theoretical prediction, margin
squeeze regulation therefore does not seem to represent a suitable safeguard for effective competition nor a beneficiary instrument for consumers in the case of a wholesale
monopoly when an integrated competitor is present. Second, in the case of wholesale competition (Table 5.5), margin squeeze regulation instead significantly reduces the
wholesale market price, which resembles the net effect of the general margin squeeze
impact, MSR, and the interaction effect WC x MSR reported in Table 5.4. The pairwise
comparison likewise confirms the positive and significant impact on the retailer’s profit
compared to the unregulated regime. Again, there is no significant negative impact on
the retail market price. Accordingly, the effect on consumer welfare is also insignificant.
Therefore, it is concluded that the decline of the wholesale market price that results
from margin squeeze regulation allows the retailer to increase its profitability, but retail
prices do not decrease proportionately and hence, consumers are not better off. Third,
in the case of wholesale competition with price commitment, a positive and significant
effect on the retail market price (Table 5.6) advises further skepticism with regard to
margin squeeze regulation and its impact on consumers. The magnitude of the relative price increase is estimated at 17.08 pp. The price increase benefits the integrated
firms by means of significantly higher profits, whereas the effect on the downstream retailer’s profit is insignificant. In sum, a margin squeeze regulation is clearly detrimental
144
5.3 Results
F IGURE 5.3: Median wholesale (dashed) and retail (solid) market prices of students and experts.
80
100
Experts (WCPC-NR)
0
20
40
60
80
60
40
20
0
Median market price
100
Students (WCPC-NR)
0
300
600
900
1200
1500
1800
0
100
200
300
400
500
600
Time in seconds
to consumers’ interest in this scenario as consumer surplus decreases significantly by
16.53 pp and may therefore even offset the gains from wholesale competition with price
commitment. For completeness, a summary of all pairwise comparisons between the
treatments by means of quantile regressions is given in Appendix B.2.
5.3.2 Validation study
Figure 5.3 illustrates the median wholesale and retail prices under WCPC-NR both
for the students treatment (left-hand panel) and for the experts treatment (right-hand
panel). While wholesale market prices of experts are lower according to the median
Experts
value over all periods (aStudents
= 43.043, am
m
Experts
Students = 50.326, y
most identical (ym
m
= 29.029), retail market prices are al-
= 50.613). Note that for both subject pools
wholesale prices are different from zero which is the theoretical prediction.
First, market outcomes between experts and students are compared based on the entire
time horizon of the experiment. In particular, the null hypothesis is that the median
market prices in the students sample and the median market prices in the experts sample are from populations with the same distribution. According to a Mann-Whitney U
test, there is no significant difference in wholesale market prices (z = 1.42, p = 0.155) nor
in retail market prices (z = 0.71, p = 0.477). Also with respect to overall medians, i.e., the
median of market cohort medians, Fisher’s exact test does not reject the equality of me-
145
Chapter 5 Wholesale Competition and Open Access Regulation
TABLE 5.7: Quantile regressions of market outcomes on subject type in case of wholesale competition with price commitment.
Baseline: Wholesale Competition with Price Commitment & No Regulation
(WCPC-NR) with Students.
(1)
am
(2)
ym
(3)
p AB
Experts
14.465
(10.259)
1.170
(10.908)
0.613
(2.923)
Period
0.001
(0.011)
0.003
(0.008)
0.001
(0.002)
Constant
42.297⇤⇤⇤
(9.791)
48.270⇤⇤⇤
(6.920)
11.974⇤⇤⇤
(2.179)
2.043⇤⇤⇤
(0.427)
0.477⇤⇤⇤
(0.078)
Observations
21,618
21,618
21,618
21,618
21,618
Covariate
(4)
pD
1.839⇤⇤⇤
(0.600)
< 0.001
(< 0.001)
(5)
f
CS
0.003
(0.126)
> 0.001
(< 0.001)
Clustered standard errors (by market cohorts) in parentheses.
⇤ p < 0.10, ⇤⇤ p < 0.05, ⇤⇤⇤ p < 0.01
dian market prices at the wholesale level (p = 0.637) or the retail level (p = 1.0). Finally,
the same result is obtained by a quantile regression that investigates the differences of
the subject pools while controlling for the time trend and intra-cluster correlation. In
order to obtain a comparable data basis with an equivalent number of periods for the
experts and students treatments, the measures of the students treatments are averaged
over three subsequent 500 ms intervals. As shown in Table 5.7, the effect of the expert subject pool is insignificant for all market variables except the retailer’s profit. The
higher profit of the entrant can be attributed to a larger spread between wholesale and
retail prices in a subset of individual market cohorts in the experts treatment, which is
also indicated by the negative coefficient for the median wholesale market price.
Naturally, general and conclusive evidence cannot be derived based on findings of statistical insignificance. However, in addition to the finding of statistical indifference,
descriptive measures as portrayed in Figure 5.3 show quantitatively similar and qualitatively equal behavior for both subject pools.
146
5.4 Repeated game analysis of tacit collusion incentives
5.4 Repeated game analysis of tacit collusion incentives
In an effort to relate the empirical findings of the experiment to the four timing models introduced in Section 5.2, their theoretical predictions are considered in a repeated
game context. Thereby, a comparison to observed experimental results may reveal
which of the timing models best captures endogenous timing under the continuous
time framework. Considering the benchmark scenario of an unregulated wholesale
monopoly, observed wholesale prices suggest that the wholesale provider does generally not foreclose the downstream retailer, but rather charges the monopolistic wholesale price. Moreover, there is no evidence that margin squeeze regulation reduces
wholesale prices in the monopoly scenario. Both observations are in line with predictions by Timing Models (2) and (4) and contradict predictions by Timing Models (1)
and (3). This may be considered as support for the experimental design as it is in line
with the intention to model the non-integrated retailer as a competitive fringe, whose
reaction is immediate, but subsequent and anticipated by the integrated firms. Furthermore, median prices for all wholesale competition treatments are significantly above
the competitive outcome, which is an equilibrium in all timing models. More specific, the significant increase in wholesale and retail prices from wholesale monopoly
to wholesale competition contradicts the consensus prediction. Whereas wholesale
prices close to amax = 100 may be interpreted as an indication for the foreclosure outcome, which is predicted by Timing Models (1) and (3), observed retail prices close to
pmax = 100 are in line with the JPM outcome, but diverge from predicted retail prices
in the foreclosure outcome. This suggests the presence of substantial tacit collusion
among integrated firms in the wholesale competition scenario.
Although experiments in continuous time have thus far been primarily used to consider static one-shot games (Friedman and Oprea, 2012; Bigoni et al., 2015; Kephart and
Friedman, 2015), continuous time may also be interpreted as an infinite repetition of a
one-shot game as described by the timing models. In this context, the incentives to tacitly collude in the upstream market can be compared in the spirit of Nocke and White
(2007) and Normann (2009) with respect to the critical discount factor that is required
147
Chapter 5 Wholesale Competition and Open Access Regulation
TABLE 5.8: Profits and discount factors under wholesale monopoly.
Firm A
Timing Model
JPM
pA
Dev
pA
Punish
pA
(1)
(2)
(3)
(4)
34.00
34.00
34.00
34.00
35.02
35.02
35.02
35.02
15.04
15.61
15.04
15.07
Firm B
dA
JPM
pB
p BDev
p BPunish
dB
0.05
0.05
0.05
0.05
17.00
17.00
17.00
17.00
25.40
25.40
25.40
25.40
15.04
14.05
15.04
11.86
0.81
0.74
0.81
0.62
TABLE 5.9: Profits and discount factors under wholesale competition.
Firm A
Timing Model
JPM
pA
Dev
pA
Punish
pA
(1), (2), (3), (4)
25.50
30.21
5.79
Firm B
dA
JPM
pB
p BDev
p BPunish
dB
0.19
25.50
30.21
5.79
0.19
to sustain collusive outcomes. Assuming a grim trigger strategy, deviations from JPM
prices amax and pmax are punished by infinite play of the competitive Nash equilibrium
(cf. Nocke and White, 2007). Individual discount factors that support collusive behavior are computed by di =
piDev piJPM
p Dev piPunish
for i 2 { A, B}, where p JPM is the firm’s share of the
JPM profit, p Dev is the maximum deviation profit that a firm can achieve by unilateral
deviation, and p Punish is the firm’s profit in periods after deviation (cf. Normann, 2009).
The minimum critical discount factor is then given by d = max {dA , dB }.
Following this approach, the minimum critical discount factors can be computed for
wholesale monopoly and competition. Table 5.8 denotes the respective profits and discount factors for both integrated firms in case of a wholesale monopoly for each timing
model. Likewise, Table 5.9 states profits and discount factors under wholesale competition. Here, critical discount factors are identical across all timing models, because in
each model punishment is exercised through the competitive equilibrium. Moreover,
integrated firms’ critical discount factors under wholesale competition are symmetric
because collusive and deviation profits are calculated as expected values, i.e., firms expect to be the access provider with probability one half.14
14 Alternatively,
one may assume that firms gain the entire wholesale profit if they deviate. Irrespective,
the ensuing minimum critical discount factor d = 0.43 is still lower than the ones reported in Table 5.8.
148
5.4 Repeated game analysis of tacit collusion incentives
Pairwise comparisons of minimum critical discount factors under wholesale monopoly
and competition show that collusion is sustainable for a larger range of discount factors under wholesale competition, independent of the assumed timing of the one-shot
game. More specifically, Firm B has a stronger incentive to deviate in the case of a
wholesale monopoly, because foregone profits in the case of punishment are relatively
low compared to its JPM profit share. In contrast, in the case of wholesale competition, expected JPM profits are higher, while profits in the case of punishment are lower,
thus making a deviation less attractive. Therefore, tacit collusion is less likely in the
wholesale monopoly setting than in the wholesale duopoly setting. This may provide
a theoretical rationale for Result 1.
However, notice that this does not provide a rationale for Result 2, because the same
theoretical analysis applies to the case of wholesale competition with price commitment. To see this, consider price commitment to induce sequential-move rather than
simultaneous-move interaction between the integrated firms regarding the wholesale
price.15 Evidently, this does not apply to the wholesale monopoly scenario and it is easy
to see that this would also not change the equilibrium in the wholesale competition scenario: As each of the integrated firms has an incentive to be the access provider, the first
mover will anticipate to be undercut by the second mover and thus set the minimum
feasible access price, just like when access prices are determined simultaneously. Consequently, the alternative timing would result in the same critical discount factors and
therefore the same prediction with respect to the incentives for tacit collusion. From
a more behavioral perspective, one could argue that the price commitment limits the
extent to which one of the integrated firms can immediately retaliate the other (in the
sense of the grim trigger strategy), which therefore makes the punishment less severe,
and ultimately tacit collusion less likely. However, in an infinitely repeated game this
lack of punishment in a short (finite) period does not matter.16 But from a behavioral
15 Note
that this timing makes sense only in Timing Models (1) and (2), because it is the very nature of
Timing Models (3) and (4) that integrated firms’ decisions are made simultaneously in the upstream
and downstream markets.
16 Obviously, it would matter in a finitely repeated game. However, note that in this case the only
subgame-perfect equilibrium would also be to play the (unique) equilibrium of the one-shot game in
each period. That is, for the case of wholesale competition, and irrespective of a price commitment, the
149
Chapter 5 Wholesale Competition and Open Access Regulation
perspective it may. After all the price commitment is able to secure the second-mover
a guaranteed wholesale profit for a (short) period of time and as such, it may stimulate
a notion of competition for the market that—in line with Result 2—amplifies the competitive process.
5.5 Discussion
Although the regulation of access to an essential upstream resource is a perennial issue for policymakers and industry stakeholders, the competitive supply of the bottleneck resource by vertically integrated firms is investigated only recently in the theoretical economic literature. By means of an economic laboratory experiment this study
scrutinizes these theoretical analyses, particularly with respect to the effectiveness of
wholesale competition in the relevant case when there are only two access providers.
The results indicate that wholesale duopoly markets may be severely affected by high
levels of tacit collusion, such that market performance in terms of market prices and
consumer surplus is worse than with a single access provider (Result 5.1). In particular, this is found to be the case under standard homogeneous Bertrand competition at
the wholesale level, which is frequently assumed in the theoretical investigations (e.g.,
Bourreau et al., 2011, 2015). In this vein and in the spirit of a more behaviorally oriented
regulation (Normann and Ricciuti, 2009; Lunn, 2014), this experimental analysis serves
as a regulatory testbed, which points at possible behavioral issues that may arise in
practice. After all, in light of the tremendous impact that regulatory decisions have on
the respective industry and—especially in the case of network industries—also on other
industries, policymakers should be particularly mindful when theoretical predictions
are not confirmed in the laboratory. In the present context, a simple price commitment rule significantly improves market outcomes, although the competitive intensity
in the wholesale market remains below the theoretical prediction (Result 5.2). Furthermore, in reference to the theoretical analyses by Petulowa and Saavedra (2014) and
competitive outcome would be played. Consequently this model variant does not provide a theoretical
rationale for Result 2 either.
150
5.5 Discussion
Jullien et al. (2014), the experimental results give a clear indication regarding the theoretically ambiguous effect of margin squeeze regulation on retail prices in the presence
of wholesale competition by vertically integrated firms. More specifically, the experimental evidence supports the rationale that the ban of a margin squeeze can impede
the intensity of competition in the retail market (Result 5.3). Moreover, the experiment
points to a particular problem of applying the margin squeeze rule to an environment
of multiple firms operating in the wholesale and retail market: When tacit collusion in
the wholesale market is stable and leads to prices above the Nash equilibrium, retail
pricing is constrained correspondingly. Especially the integrated firm, which naturally
has an incentive to be more aggressive in the retail market because it is not affected by
the softening effect, may be restricted in setting lower retail prices as long as it decides
not to undercut prices in the wholesale market. Although the margin squeeze rule,
as an implicit Open Access rule, ensures non-discrimination between competitors, the
premise to treat all market participants alike is not aligned with the diverse incentives
that occur in the case of simultaneous retail and wholesale competition, e.g., due to the
consideration of opportunity costs by the access provider. Thus, non-discrimination of
competitors may not always be in the best interest of the consumer.17
With respect to the limitations of the experimental study, it should be noted that firms’
investment incentives are not considered under the various market scenarios and thus,
experimental insights are constrained to short-term issues of static efficiency. However,
in many industries, particularly in network industries such as telecommunications, dynamic efficiency is considered to be at least equally important by policymakers. Nevertheless, the findings may still be informative in this context, as there is generally an
inherent trade-off between static and dynamic efficiency (see Chapter 2) with some
notable exceptions (Klumpp and Su, 2010). That is, dynamic investment incentives
are to a large extent influenced by the expectations about the future (static) benefits
that arise from a given market structure (especially the market shares of competitors,
see Klumpp and Su, 2015), the obtained results may inform further research regarding
17 Note
that an additional well-known negative effect of non-discrimination on competition is articulated
by the theory of restoring monopoly power (Rey and Tirole, 2007b), where non-discrimination allows the
upstream firm to resolve its commitment problem.
151
Chapter 5 Wholesale Competition and Open Access Regulation
the effects that arise under infrastructure-based competition with multiple wholesale
providers. In this context, the experimental results also cast doubt on the premise that
infrastructure-based competition should be the undisputed regulatory goal (Cave and
Vogelsang, 2003; Cave, 2006a), particularly when Open Access for independent retailer
is required at the same time (e.g., as indicated by the European Commission’s (2014b)
digital agenda). This is in line with empirical findings by Höffler (2007) which suggest
that infrastructure duplication costs may be higher than the gains from (supposedly)
intensified competition. More general with respect to economic experimentation, a second concern arises with respect to the external validity of findings, although the validation study with industry professionals corroborates the robustness of the obtained
student subject pool results. Naturally, experimental results do not directly carry over
to actual markets, however, at the same time, one should also be cautious to believe
that theoretical predictions will hold in practice when they already fail in a laboratory
environment. Furthermore, note that the results are based on the relative differences
between treatments and should thus not be affected by factors that are held constant
across treatments. Nevertheless, an empirical field study of wholesale access in context
of infrastructure competition would certainly represent a highly valuable contribution
complementing theoretical and experimental work.
Finally, this study may also inspire future work. First, rules and remedies that are
deemed to intensify competition at the wholesale level could be investigated in further
depth. While two alternative modes of wholesale competition which yield market outcomes below and above the wholesale monopoly treatment have been considered, the
investigation of the underlying competitive process and further investigation of instruments that may intensify competition at the wholesale level appear promising. Second,
the presented analysis may be extended by a variation of the number of competitors
and retailers as well as the introduction of asymmetry between the integrated firms.
With regard to the competition across different access levels and quality layers—as in
Internet and telecommunications markets—such an extension could provide valuable
insight for decision makers and regulators within these fast-moving industries.
152
Chapter 6
Number Effects and Tacit Collusion in
Oligopolies
I
N oligopolistic markets, concerns about market power and coordinated behavior
frequently confront competition and regulatory authorities with the question: How
many competitors are enough to ensure competition? Thereby, it is of particular interest how
the competitiveness of a market is affected if a low number of competitors is further reduced through consolidation, i.e., if the number of firms is reduced from four to three
or from three to two. For example, several high-profile merger control proceedings in
the European Union1 as well as in the US2 have dealt with cases that would reduce the
remaining number of competitors from four to three major mobile telecommunications
operators in the respective relevant market. In the US airline industry, the Department
of Justice had initially filed a lawsuit to block the merger between American Airlines
and US Airways that reduced the number of legacy carriers from four to three, explicitly
referring to the low number of competitors as a critical threat to effective competition
(Stewart, 2013). Even in a high-tech commodity industry like the hard disk drive industry, consolidation among manufacturers raises the question whether there is a magThis chapter is based on joint work with Niklas Horstmann and Jan Krämer (Horstmann et al., 2016b).
3G Austria / Orange Austria (European Commission, 2012), Telefónica Deutschland / EPlus (European Commission, 2014c), Hutchinson 3G UK / Telefónica Ireland (European Commission,
2014d).
2 AT&T / T-Mobile US (Federal Communications Commission, 2011), Sprint Corp / T-Mobile US (Federal
Communications Commission, 2014).
1 Hutchinson
153
Chapter 6 Number Effects and Tacit Collusion in Oligopolies
ical number to reconcile scale synergies and pro-competitive effects (Igami and Uetake,
2015). Similar to competition authorities, sector-specific regulatory agencies implicitly
or explicitly examine the sufficient number of competitors when assessing the need for
ex ante access regulation. For instance, geographically segmented deregulation of the
wholesale broadband access market in the UK is conditioned on the number of active
competitors in a region (Ofcom, 2014). Likewise, in media retail markets, which in
some countries are characterized by repeated direct public intervention, competition
authorities are required to determine the sufficient number of local sellers to achieve
sustainable coverage in combination with competitive prices (Balmer, 2013).
In general, the number of firms in a specific market is determined endogenously by
the competitive process and particularly by firms’ entry and exit decisions. However,
as described above, in merger cases and regulatory proceedings, authorities are often
required to determine a specific number of competitors exogenously. This makes it necessary to estimate the impact of number effects on the competitiveness in that market.
Obviously, markets in practice exhibit many idiosyncrasies that impact competitiveness
and require a case-by-case analysis before a merger can be cleared or before regulatory
remedies are imposed or lifted. Yet, a general relationship between the number of competitors and the competitiveness of a given market, above and beyond any market peculiarities, is frequently assumed. To this end, it is well-known that equilibrium predictions for market prices (quantities) are generally decreasing (increasing) with a higher
number of competitors, i.e., markets become more competitive. However, the impact
on the degree of tacit collusion, i.e., the ability of firms to sustain a supra-competitive
outcome above (below) the equilibrium, is not as clear.
Based on three complementary studies this chapter investigates the research hypothesis that tacit collusion in oligopolistic markets with two, three, and four competitors
decreases strictly monotonically with the number of competing firms. From a methodological point of view, experimental laboratory experiments are well suited to address
this question, because they allow to observe out-of-equilibrium behavior while controlling for environmental conditions. In this context, it has previously been concluded that
154
tacit collusion is “frequently observed with two sellers, rarely in markets with three
sellers, and almost never in markets with four or more sellers” (Potters and Suetens,
2013, p.17). Surprisingly, a review of the extant literature shows that there is actually
no robust empirical evidence that would support the claim of a strictly monotonically
decreasing relationship between the number of firms and the degree of tacit collusion
in a given market. Whereas, a meta-analysis of the extant literature supports the notion that duopolies are significantly more prone to tacit collusion than quadropolies,
i.e., that “two are few and four are many” (Huck et al., 2004b, p.435), there is no empirical support for a significant effect when moving from four to three firms. However,
the lack of statistical power across and within existing studies precludes a conclusive
evaluation of a strictly monotonic relationship between the number of firms and the
degree of tacit collusion in a market. Moreover, the review of the extant literature reveals a lack of systematic evaluation of such number effects under different competition
models (Cournot vs. Bertrand), symmetric and asymmetric firms, and under consideration of different theoretical equilibrium predictions (Nash vs. Walras). Therefore, two
laboratory experiments are conducted in this chapter, which are explicitly designed to
systematically test for number effects on tacit collusion under price and quantity competition, as well as with symmetric and asymmetric firms. Thereby, the findings show
a significant competitive effect from four to three firms as well as from three to two firms.
In fact, the empirically observed decrease in the degree of tacit collusion is almost identical from four to three as from three to two, suggesting a linear number effect for highly
concentrated oligopolies with regard to the (in)ability to coordinate above the theoretical Nash prediction.
The remainder of this chapter is organized as follows. Next, Section 6.1 reviews the
extant experimental literature and conducts a meta-analysis. In Sections 6.2 and 6.3
the design and results of the experiments with symmetric and asymmetric firms are
reported, respectively. Section 6.4 discusses the findings pooled over all three studies
and highlights their policy implications for both antitrust and regulatory authorities.
155
Chapter 6 Number Effects and Tacit Collusion in Oligopolies
6.1 Review and meta-analysis of the experimental literature
In order to investigate the general relationship between the number of competitors
and the competitiveness in a given market, economic laboratory experiments are particularly suited, because they allow to identify systemic effects and to isolate distinct
sources for tacit collusion through controlled variation of exogenous variables (see Section 3.2). In particular, experiments allow to isolate the effect of the number of competitors on firms’ ability to coordinate through randomization, while holding constant
any potential confounding variable. By this means, the experimental method avoids
endogeneity concerns (e.g., of the observed number of competitors) inherent to field
data (Angrist and Pischke, 2010), which are especially pronounced in the context of the
structure-conduct-performance paradigm (Scherer and Ross, 1990; Schmalensee, 1990)
of industrial organization. Moreover, empirical field studies are naturally framed in a
specific market context and are thus neither generalizable per se nor directly applicable
to other market scenarios as causal relationships are inherently difficult to prove (see
Einav and Levin, 2010, for a discussion of generalizability of empirical industry studies). Particularly with regard to the issue of tacit collusion, which is notoriously hard
to detect in field studies, laboratory experiments can provide general insights by analyzing in- and out-of-equilibrium strategies and respective market outcomes relative to
benchmark equilibria predicted by economic theory.
Consequently, it is not surprising that there are several experimental studies that investigate the drivers and impediments of tacit collusion in oligopolies. In their overview
on these experiments, Potters and Suetens (2013) conclude, among others, that “the
scope of collusion is strongly affected by the number of competitors” (p.17). With respect to the effect of the number of competitors on tacit collusion, Potters and Suetens
suggest a strictly monotonic relationship, referring to individual experimental studies
that have employed posted-offer markets, Bertrand competition or Cournot competition. Although several of these studies find that tacit collusion is generally lower for
a larger number of competitors, e.g., in markets with four relative to two competitors
(e.g., Huck et al., 2004b; Orzen, 2008), as described above, evidence for the stipulated
156
6.1 Review and meta-analysis of the experimental literature
monotonic number effect is rather scarce, in particular with regard to the assessment of
tacit collusion in markets with three and four competitors.
6.1.1 Experimental designs
Most oligopoly experiments implement one of the two workhorse models in industrial organization: price competition à la Bertrand (Fouraker and Siegel, 1963; Dolbear
et al., 1968; Dufwenberg and Gneezy, 2000; Orzen, 2008; Davis, 2009; Fonseca and Normann, 2012) or quantity competition à la Cournot (Fouraker and Siegel, 1963; BoschDomènech and Vriend, 2003; Huck et al., 2004b; Waichman et al., 2014).3 A third strand
of literature observes tacit collusion in posted-offer markets, i.e., simultaneous competition in prices and quantities (Ketcham et al., 1984; Alger, 1987; Brandts and Guillén, 2007; Ewing and Kruse, 2010). As the latter experiments use very diverse models
and are hence hardly comparable to one another, the focus here is on price or quantity
competition. Table 6.1 lists the ten oligopoly experiments which are surveyed in this
meta-analysis and which all vary the number of competitors n in a market in one way
or another.4
Six experiments employ price competition. Four of those investigate homogeneous
Bertrand competition, i.e., firms’ products are perfect substitutes. The remaining two
experiments use differentiated price competition, i.e., competitors’ products are differentiated with regard to quality or consumers have heterogeneous preferences: Dolbear
et al. (1968) consider a model in which the cross-price elasticity is half the own-price
elasticity and Orzen (2008) models a fraction of consumers to be price-insensitive “convenience shoppers” (Orzen, 2008, p.392). All of the four quantity competition experiments included in this meta-analysis employ a homogeneous Cournot model.
3 Note
that merger experiments induce asymmetry exogenously (see Götte and Schmutzler, 2009, for
a comprehensive review) or endogenize merger formation which yields asymmetric markets postmerger (Lindqvist and Stennek, 2005). In order to prevent path dependencies from merger formation,
only data from those experimental studies that vary the number of competing firms exogenously across
treatments is used for this meta-analysis.
4 To the extent of the author’s knowledge, the list in Table 6.1 is complete with the exception of Abbink
and Brandts (2005, 2008) for which no complete experimental data was attainable.
157
Chapter 6 Number Effects and Tacit Collusion in Oligopolies
TABLE 6.1: Economic laboratory experiments that vary the number of competing firms.
Information
Study
Competition
Fouraker and Siegel (1963)
Dolbear et al. (1968)
Dufwenberg and Gneezy (2000)
Orzen (2008)
Davis (2009)
Fonseca and Normann (2012)
Homogeneous
Differentiated
Homogeneous
Differentiated
Homogeneous
Homogeneous
Complete
Perfect
Matching
n
3/ 7
3
3
3
3
3
Partner
Partner
Stranger
Partner, Stranger
Partner
Partner
{2, 3}
{2, 4, 16}
{2, 3, 4}
{2, 4}
{2, 3, 4}
{2, 4, 6, 8}
3/ 7
3
3/ 7
3/ 7
Partner
Partner
Partner
Partner
{2, 3}
{2, 3}
{2, 3, 4, 5}
{2, 3}
Bertrand (price) competition
3
3/ 7
3
3
3/ 7
3
Cournot (quantity) competition
Fouraker and Siegel (1963)
Bosch-Domènech and Vriend (2003)
Huck et al. (2004b)
Waichman et al. (2014)
Homogeneous
Homogeneous
Homogeneous
Homogeneous
3
3/ 7
3
3
3: applicable | 7: not applicable | 3/ 7: both (as treatment variable)
Experiments differ further in the amount of information provided to participants. In
a situation of complete information, each firm, represented by an individual participant, knows about (or can retrieve) the cost and demand function of all firms in the
market. Moreover, a firm with perfect information can observe all decisions made by
its competitors, and hence, has knowledge over the full history of the game. Lastly, all
but one study employ a fixed matching of firms over the entire time horizon. Instead,
Dufwenberg and Gneezy (2000) match firms randomly in each period. Orzen (2008)
additionally compares partner and stranger matching in a between-subject manner.
6.1.2 Measuring competitiveness as the degree of tacit collusion
In order to compare number effects on competitiveness or likewise, tacit collusion,
across heterogeneous data sets from different experimental designs, a uniform performance criterion is required. As absolute price or quantity levels are inconclusive across
experiments, different metrics are proposed in the extant literature to measure competitiveness in experimental oligopoly outcomes. For a review of Cournot experiments,
Huck et al. (2004b) report the ratio between a market’s average total quantity Q and the
total Nash quantity Q Nash , r = Q/Q Nash . However, as Engel (2007) points out, r is “sensitive to arbitrary changes in the level of Q N [ash] ” (p.494). In addition, the measure is
not well suited to quantify and compare non-equilibrium outcomes between treatments
158
6.1 Review and meta-analysis of the experimental literature
and experimental designs, because it does not incorporate the joint profit maximizing
(JPM) outcome as a second benchmark.
Therefore, the measure introduced in Subsection 3.3.2, which has previously been employed in meta-analyses by Engel (2007) and by Suetens and Potters (2007), is utilized
here as well. Thus, tacit collusion is measured as the relative deviation of average price
from the theoretical equilibrium E 2 { Nash, Walras} towards the JPM price p JPM . Formally,
j E = j Ep =
p
p JPM
pE
.
pE
In this vein, j E represents the degree of tacit collusion based on prices as compared to either the Nash equilibrium or the Walrasian (competitive) equilibrium as the theoretical
prediction. The Walrasian equilibrium assumes all competitors to be price-takers and
thus, under homogeneous Bertrand competition the Nash prediction and the Walrasian
prediction coincide. Moreover, under some regularity conditions, Walrasian prices cannot exceed Nash prices in any oligopoly competition model, i.e., pWalras  p Nash . If
j E = 0, the average market price p corresponds to the theoretical prediction by the
equilibrium concept E. If j E = 1, the market is completely collusive and competitors
behave like in the case of a monopoly. Note that j E may exceed one if joint profit is
not monotonic in prices; the measures’ lower limits, however, depend on the experimental design. Suetens and Potters (2007) employ the same collusion metric based on
Nash predictions to investigate relative differences of tacit collusion under Bertrand
and Cournot competition.5
In addition, Friedman (1971) suggests a theoretical benchmark to assess the likelihood
“that tacit collusion can be sustained as an equilibrium in an infinitely repeated game
context as part of a grim trigger strategy” (Suetens and Potters, 2007, p.73), which is
given by Friedman =
P JPM P Nash
P De f ect P JPM
with P De f ect as the maximum profit for a firm that
unilaterally deviates from a collusive agreement. Hence, the Friedman index measures
the incentive to tacitly collude by comparing the collusive markup on the Nash profit
5 Suetens
and Potters (2007) exclude negative prices in Cournot experiments from their calculation of the
degree of tacit collusion. In this meta-analysis, however, negative prices are considered as well, as they
correctly reflect the high competitiveness of excess capacity in Cournot markets.
159
Chapter 6 Number Effects and Tacit Collusion in Oligopolies
to the additional profit for defecting from cooperation. In repeated oligopoly experiments each firm has to trade off short-term profits from deviating to foregone profits in
future periods. The higher the Friedman index, the less profitable is a deviation from
a collusive agreement.6 Although the Friedman index assumes an infinitely repeated
game, it may nonetheless be informative in the context of finitely repeated games in experiments with fixed lengths across treatments as it is well-known that tacit collusion
is no phenomenon that is limited to experiments with random termination rules.
6.1.3 Results of the meta-analysis
Table 6.2 reports the number of independent observations N, the two collusion metrics
j E , and the Friedman index for all experiments and treatments considered in this metaanalysis.7 The following analysis is carried out in two steps: At first number effects
are examined only within a single study (intra-study). Subsequently, tacit collusion in
duopolies, triopolies, and quadropolies is compared across all studies (inter-study).
R ESULT 6.1. Within and across the surveyed oligopoly experiments, markets with two firms
are significantly more prone to tacit collusion than markets with three as well as four firms,
everything else being equal. However, no significant difference in the degree of tacit collusion
can be found between three and four firms.
With respect to the intra-study analysis, data on the level of independent observations
could be obtained for five experiments.8 Table 6.3 provides p values from one-tailed
non-parametric Mann-Whitney U tests of intra-study number effects on tacit collusion
in these experiments. Following the hypothesis of a strictly monotonic relationship, the
null hypothesis is that tacit collusion is always higher in a market with more firms. With
6 For
Orzen (2008), the Friedman index has to be averaged over all three successive phases in each treatment in order to gain a single index value.
7 The original experimental data is either collected from tables in the respective study, downloaded from
an online repository, or provided by the authors. One exception is Bosch-Domènech and Vriend (2003)
for which the data is retrieved from figures.
8 The author thanks Hans-Theo Normann and Henrik Orzen for providing the experimental data used in
Huck et al. (2004b) and Orzen (2008), respectively.
160
TABLE 6.2: Degrees of tacit collusion in economic laboratory experiments that vary the number of competing firms.
Periods†
Study
Treatment
Fouraker and Siegel (1963)
Complete information
[1, 15] 2 [1, 15]
Incomplete information
[1, 15] 2 [1, 15]
Dolbear et al. (1968)
Complete information
[8, 12] 2 [1, 15]
Dufwenberg and Gneezy (2000)
2/3/4
[1, 10] 2 [1, 10]
Orzen (2008)
Fixed matching
[1, 90] 2 [1, 90]
Random matching
[1, 90] 2 [1, 90]
n
N
j Nash
jWalras
Friedman
2
3
2
3
2
4
2
3
4
2
4
2
3
2
3
4
2
4
6
8
17
10
17
11
18
9
12
8
6
6
6
6
6
6
6
6
6
6
6
6
0.412
0.039
0.149
0.019
0.300
0.040
0.260
0.067
0.077
0.352
0.025
0.113
0.008
0.113
0.006
0.006
0.504
0.060
0.025
0.011
0.412
0.039
0.149
0.019
0.500
0.257
0.260
0.067
0.077
0.604
0.381
0.462
0.391
0.113
0.006
0.006
0.504
0.060
0.025
0.011
0.766
0.311
0.766
0.311
1.250
1.250
1.000
0.497
0.331
0.624
0.206
0.624
0.206
0.754
0.376
0.251
1.020
0.338
0.202
0.145
2
3
2
3
2
3
2
3
2
3
2
3
4
5
2
3
2
3
16
11
16
11
9
6
9
6
9
6
6
6
6
6
12
13
10
11
0.244
0.266
0.114
0.260
0.296
0.176
0.159
0.107
0.164
0.491
0.403
0.032
0.065
0.109
0.154
0.265
0.046
0.062
0.585
0.367
0.629
0.370
0.765
0.451
0.614
0.484
0.612
0.304
0.801
0.516
0.439
0.260
0.615
0.367
0.651
0.469
1.000
0.750
1.000
0.750
0.889
0.732
0.889
0.732
0.889
0.732
0.889
0.750
0.640
0.556
0.889
0.750
0.889
0.750
Bertrand (price) competition
2np/3np/4np
[1, 220] 2 [1, 220]
Fonseca and Normann (2012)
NoTalk
Fouraker and Siegel (1963)
Complete information
[1, 22] 2 [1, 22]
Incomplete information
[1, 22] 2 [1, 22]
Easy
[1, 22] 2 [1, 22]
Hard
[1, 22] 2 [1, 22]
Hardest
[1, 22] 2 [1, 22]
Huck et al. (2004b)
Unified frame
[1, 25] 2 [1, 25]
Waichman et al. (2014)
DSNC/TSNC
[1, 17] 2 [1, 17]
DMNC/TMNC
[1, 17] 2 [1, 17]
[1, 29] 2 [1, 29]
Cournot (quantity) competition
Bosch-Domènech and Vriend (2003)
161
†
Periods used to compute the average degree of tacit collusion. If possible, data from all periods is used to maximize comparability.
6.1 Review and meta-analysis of the experimental literature
Davis (2009)
Chapter 6 Number Effects and Tacit Collusion in Oligopolies
TABLE 6.3: Intra-study one-tailed Mann-Whitney U tests and associated p values.
Study
Treatment
j Nash
n
jWalras
Bertrand (price) competition
Fouraker and Siegel (1963)
Orzen (2008)
Davis (2009)
Complete information
Incomplete information
Fixed matching
Random matching
2np/3np/4np
2 vs.
2 vs.
2 vs.
2 vs.
2 vs.
2 vs.
3 vs.
3 < 0.001 < 0.001
3
0.003
0.003
4
0.005
0.005
4
0.002
0.002
3
0.008
0.008
4
0.008
0.008
4
0.437
0.437
Cournot (quantity) competition
Fouraker and Siegel (1963)
Huck et al. (2004b)
Complete information
Incomplete information
Unified frame
2 vs.
2 vs.
2 vs.
2 vs.
3 vs.
3
3
3
4
4
0.294
0.008
0.084 < 0.001
0.019
0.004
0.019
0.002
0.261
0.261
the exception of the measure based on Nash predictions for Fouraker and Siegel’s (1963)
Cournot treatments, all test results indicate that tacit collusion is higher in duopolies
than in triopolies (2 vs. 3) or quadropolies (2 vs. 4) at the 5% level of significance.
However, triopolies are not found to be more prone to tacit collusion than quadropolies
(3 vs. 4), neither under Bertrand competition nor under Cournot competition.
For inter-study comparisons the most comparable treatments between studies are selected in an effort to rule out any other explanations for differences other than the
number of competitors. Thus, only treatments with complete and perfect information
are considered for the following analysis.9 Consequently, there are ten independent
duopoly observations, seven independent triopoly observations, and six independent
quadropoly observations. As there is only a single study for any n > 4 the statistical
analysis is limited to markets with n 2 {2, 3, 4} firms. The Friedman index, which is
suggested to assess the likelihood of tacit collusion, predicts poorly if correlated with
j Nash (r = 0.213, p = 0.330) but is positively and significantly correlated with jWalras
(r = 0.593, p = 0.003). In order to control for potential dependencies between treatments
9 The
following treatments reported in Table 6.2 are not considered in this step of the inter-study analysis: Incomplete information (Fouraker and Siegel, 1963), Random matching (Orzen, 2008), Hard and
Hardest (Bosch-Domènech and Vriend, 2003), and DMNC/TMNC in which participants are managers
instead of students (Waichman et al., 2014).
162
6.1 Review and meta-analysis of the experimental literature
from the same study, i.e., different base levels of tacit collusion between experimental
settings, the following three-level linear random-intercept model is estimated:
E
js,m,n
= b0 + x s + z m
+ b Duopoly · Duopoly
+ b Quadropoly · Quadropoly
+ b Cournot · Cournot
+ es,m,n ,
E
where js,m,n
is the average degree of tacit collusion j E of markets with n competitors
under model m 2 { Bertrand, Cournot} in study s, z m is the error component shared
between observations of the same model in study s (see Bertrand and Cournot treat-
ments in Fouraker and Siegel, 1963), and x s is the error component shared between
observations from the same study. The results, as portrayed in Table 6.4, confirm the
insight of the above intra-study findings that there is significantly more tacit collusion
in duopolies compared to triopolies and quadropolies. Furthermore, there is no significant difference in tacit collusion between triopolies and quadropolies. In particular, the
degree of tacit collusion is, on average, 26 pp higher in duopolies than triopolies according to both collusion measures. On the contrary, the same does not hold for the comparison between markets with three and markets with four firms as triopolies are found
to have, on average, an almost identical degree of tacit collusion than quadropolies.
Also notice that the regression analysis replicates the finding by Suetens and Potters
(2007) that Bertrand colludes more than Cournot—however, only if tacit collusion is based
on Nash predictions. In contrast, when compared to Walrasian equilibrium, this effect
is significant in the opposite direction. Thus, if a competitive market outcome where
price equals marginal cost represents the benchmark for the degree of tacit collusion,
Cournot may collude more than Bertrand.
All these results hold if tacit collusion metrics are averaged over all treatments from
each study with the same competition model and two, three, or four firms, respectively
(see Table B.14 in Appendix B.3 for results of the respective multilevel mixed-effects
163
Chapter 6 Number Effects and Tacit Collusion in Oligopolies
TABLE 6.4: Multilevel mixed-effects linear regressions of tacit collusion on number of competitors and competition model on the basis of most comparable treatments.
Baseline: Triopoly & Bertrand competition.
(1)
j Nash
(2)
jWalras
Duopoly
0.259⇤⇤⇤
(0.048)
0.261⇤⇤⇤
(0.031)
Quadropoly
0.020
(0.060)
0.003
(0.039)
Cournot
0.227⇤⇤
(0.088)
0.263⇤⇤⇤
(0.050)
Constant
0.056
(0.067)
0.156⇤⇤⇤
(0.049)
9
10
23
9
10
23
Covariate
Studies
Models
Observations
Baseline: Bertrand triopoly. Standard errors in parentheses.
⇤ p < 0.10, ⇤⇤ p < 0.05, ⇤⇤⇤ p < 0.01
regressions). Furthermore, these findings can also be replicated by a meta-regression, a
method vastly used in medical research (see, e.g., Higgins and Thompson, 2002), which
takes into account the reliability of sample means from different studies by controlling
for study-specific standard errors. The results of the meta-regressions are in line with
the findings of the multilevel mixed-effects linear regressions presented in Table 6.4
and yield similiar effect sizes. See Appendix B.3 for a presentation and discussion of
estimates.
Although the previous analyses control for different base levels of tacit collusion between experiments via multilevel mixed-effects regressions as well as for the reliability
of sample means via meta-regressions, the data used in the previous regression models are unbalanced with regard to the different number of independent observations
(treatments) for each number of competitors. Consequently, number effects are next
investigated inter-study also via matched samples. By this means, a comparison of n1
and n2 competitors includes all studies that have conducted treatments with n1 and n2
competitors. Note that, therefore, the number of included studies varies between pairwise comparisons, e.g., when comparing two with four and two with three competitors.
164
6.1 Review and meta-analysis of the experimental literature
TABLE 6.5: Inter-study average degrees of tacit collusion and one-tailed matched-samples
Wilcoxon signed-rank tests on the basis of most comparable treatments.
Studies
j Nash
jWalras
2 vs. 3
Duopoly
Triopoly
p value
7
7
7
0.155
0.081
0.009
0.507
0.259
0.009
2 vs. 4
Duopoly
Quadropoly
p value
6
6
6
0.322
0.024
0.014
0.464
0.203
0.014
3 vs. 4
Triopoly
Quadropoly
p value
3
3
3
0.035
0.049
0.946
0.196
0.174
0.500
Table 6.5 presents average degrees of tacit collusion and p values based on one-tailed
non-parametric Wilcoxon signed-rank tests. Again, the tested null hypothesis is that
tacit collusion is higher in markets with less firms than in markets with more firms.
Test results show that tacit collusion is significantly higher in duopolies than in triopolies (2 vs. 3) and quadropolies (2 vs. 4), respectively. However, based on all experiments that run triopolies as well as quadropolies, the former is not more prone to
tacit collusion than the latter (3 vs. 4). In fact and in stark contrast to the existence of a
strictly monotonic relationship, tacit collusion may even be slightly higher in markets
with four firms (j Nash = 0.049) than in markets with three firms (j Nash = 0.035) and
this difference is almost significant at the 5% level (N = 3, p = 0.054). Again, results are
similar if tacit collusion metrics are averaged over all treatments from each study with
the same competition model and two, three, or four firms, respectively (see Table B.15
in Appendix B.3 for results of corresponding Wilcoxon signed-rank tests).
6.1.4 Discussion of the meta-analysis
Based on the survey of the extant literature, there is no pooled evidence that would
support a general strictly monotonic relationship between the number of firms and
the degree of tacit collusion in experimental oligopolies. Moreover, the studies that
165
Chapter 6 Number Effects and Tacit Collusion in Oligopolies
have examined number effects between three and four competitors within a single
study likewise conclude that there is no significant difference between triopolies and
quadropolies. For homogeneous Bertrand competition, both Dufwenberg and Gneezy
(2000) and Davis (2009) find that experimental markets converge to the Nash equilibrium with three as well as with four competitors. For homogeneous Cournot competition, Huck et al. (2004b) find more competitive output levels with four compared to
three firms in absolute terms, however, this effect is reversed if the collusion degree is
measured relative to the Nash equilibrium. In summary, the results of the inter- and
intra-study analyses are in line and suggest that the surveyed oligopoly experiments
cannot support that markets with four firms exhibit ceteris paribus a lower degree of
tacit collusion than markets with three firms. In conclusion, this contrasts the hypothesis of a strictly monotonic relationship between the number of competitors and the
degree of tacit collusion in a market.
Although meta-analyses can provide valuable insight by evaluating robustness and
external validity of systematic effects, they also have several limitations. First, the lack
of control for all differences between studies considered in the same analysis limits the
internal validity of meta-results per definitionem. Second, in this specific meta-analysis
the number of independent observations of the pairwise comparisons is rather low,
which raises concerns about statistical power. In particular, only three studies cover
both triopoly and quadropoly treatments. Moreover, the individual studies themselves
may lack statistical power to detect a difference between three and four competitors
as the number of independent observations is often lower for markets with a larger
number of competitors.10 In addition, both experiments that examine number effects
in the context of price competition in a single study employ a homogeneous Bertrand
model, which arguably represents a special case with regard to number effects, because
the theoretical prediction does in fact not change with the number of competitors. Last
but foremost, none of the experiments in the meta-study employs treatments with all
10 The three studies that examine triopolies and quadropolies (Dufwenberg and Gneezy, 2000; Huck et al.,
2004b; Davis, 2009) each base their analysis on six independent quadropoly observations.
166
6.2 Experiment with symmetric firms
the relevant characteristics considered here, i.e., Bertrand and Cournot markets with
two, three, and four firms.
6.2 Experiment with symmetric firms
Due to the lack of comprehensive evidence on a strictly monotonic relationship between the number of firms and the degree of tacit collusion in the extant experimental
literature, two oligopoly experiments are conducted based on a design that exploits the
duality between Bertrand and Cournot competition. First, this section considers the
case of symmetric firms, which is also assumed in all of the oligopoly experiments surveyed in the meta-study. Next, the subsequent section considers asymmetry between
firms.
6.2.1 Experimental design
Price competition à la Bertrand and quantity competition à la Cournot serve as good
proxies for a large share of models on oligopoly competition. As homogeneous price
competition is often deemed unrealistic and yields a discontinuous demand function,
the model by Singh and Vives (1984) is considered that generalizes the Hotelling (1929)
model to exploit the duality between price and quantity competition in differentiated
goods. More precisely, the model’s generalization to more than two firms (see Häckner, 2000) is used, which is described in further detail in Subsection 3.3.2. Appendix A.1
provides a thorough analysis of the general model with asymmetric firms and a formal
derivation of the theoretical predictions, specifically, the Nash and the Walrasian equilibria as well as the joint profit maximizing outcome. For the assessment of symmetric
firms, it is assumed that firms’ goods are differentiated horizontally but are homogeneous in vertical quality and have identical demand elasticity.
167
Chapter 6 Number Effects and Tacit Collusion in Oligopolies
Specifically, treatments are examined with Bertrand and Cournot competition in
duopolies, triopolies, and quadropolies in a full-factorial design, resulting in a total
of six treatments. In the following, these treatments are referred to with abbreviations
such as B4 for the Bertrand quadropoly treatment. The model is parametrized with
w = 100, l = 1, and q = 23 . Consequently, G =
300
2n+1 ,
L=
6n 3
2n+1 ,
and Q =
6n 6
2n+1 .
The theo-
retical benchmarks of the one-shot game for each treatment are reported in Table A.1 in
Appendix A.1. As Nash prices, quantities, and profits do not coincide under Bertrand
and Cournot competition and are additionally dependent on the number of competitors n, these values are not adequate to compare cooperative intentions, i.e., tacit collusion, across treatments. Thus, the same measure as for the meta-analysis is utilized,
i.e., the degree of tacit collusion j E .
In a further effort to maximize comparability between treatments and to prevent any
source for behavioral effects other than the treatment, input and output variables are
scaled in the following way. The action space of prices in Bertrand treatments and
quantities in Cournot treatments is equally set to [0, 100] with a minimum increment
of one and the JPM action at a price or quantity of 50. This ensures that the collusive
action is not more or less behaviorally attractive across treatments and that the search
costs of finding the collusive action are the same in all treatments. With a similar intention profits are scaled so that they would be equal in Nash equilibrium. Thereby, a
subject playing the Nash equilibrium of the one-shot game—given that its competitors
play Nash as well—would make identical profits in all treatments.11 Altogether, this
precludes confounding effects of the experimental design and parametrization.
Due to the normalization of input and output variables of the model, the different
measures of the degree of tacit collusion have two desirable characteristics in the experiment. First, the Nash prediction-based degree j Nash serves as a good predictor of
relative differences in tacit collusion between treatments as Nash equilibria vary with
the competition model as well as with the number of firms in the market. Second,
11 Alternatively,
profits may be standardized with respect to the collusive outcome. However, this would
in turn lead to different Nash profits across treatments. Hence, firms would face diverse incentives to
deviate from the theoretical Nash prediction.
168
6.2 Experiment with symmetric firms
TABLE 6.6: Nash predictions p Nash /q Nash as measured by the Walrasian-based degrees of tacit
collusion jWalras .
Duopoly
Triopoly
Quadropoly
Bertrand
Cournot
0.50
0.33
0.25
0.75
0.60
0.50
the Walrasian-based measure of tacit collusion jWalras assesses absolute differences to
a uniform baseline, as the experiment is specifically designed to have a constant Walrasian equilibrium and collusive equilibrium across treatments. Due to the normalization of input variables, choosing a price or quantity of p, q 2 [0, 100] in the experiment
directly translates to a Walrasian-based degree of tacit collusion of 2p% in the Bertrand
or 2(100
q)% in the Cournot treatments, respectively. Consequently, as the equilib-
rium price level is monotonically decreasing with the number of firms, the Walrasian
prediction associated with each treatment’s Nash equilibrium is also strictly monotonically decreasing as shown in Table 6.6. Therefore, if participants in the experiment
behave in line with the Nash prediction and do not have an inexplicable preference towards a certain integer within the interval [0, 100] or even choose prices and quantities
randomly, the Walrasian-based tacit collusion measure is expected to decrease with the
number of firms.
The consideration of the degree of tacit collusion based on the Walrasian equilibrium in
the experiment is thus not only done for completeness, but also serves two additional
purposes. First, the measure serves as a means to check whether subjects’ behavior
in the experiment are in line with decreasing Nash predictions for a larger number of
firms. Although, Nash-consistent behavior is also indicated by j Nash = 0 across markets with a varying number of firms, this equality is difficult to verify empirically as a
non-significant difference may also result from a lack of statistical power. Precisely in
this case, the Walrasian collusion measure should indeed decrease with the number of
firms, and should thus differ significantly between markets with a varying number of
firms. Second, in the model consumer surplus as well as total welfare are monotonically
169
Chapter 6 Number Effects and Tacit Collusion in Oligopolies
decreasing in prices if goods are substitutes and hence, for regulatory authorities, the
Walrasian equilibrium may constitute an additional relevant theoretical benchmark.
Perfect information is ensured in all treatments, i.e., subjects are provided with individual feedback about each competitor’s price, quantity, and profit. In an effort to prevent
that treatments evoke only short-term effects, the one-shot game is repeated 60 times.
These repetitions are referred to as periods.
6.2.2 Procedures
The experiment is computerized with the Java-based experimental software Brownie
(Müller et al., 2014). All sessions were run at the Karlsruhe Institute of Technology
in Karlsruhe, Germany in October 2014 (duopoly and triopoly sessions), April 2015
(quadropoly sessions), and May 2016 (additional quadropoly sessions). Disregarding
the first period, in which subjects familiarized themselves with the experimental software and decided on their initial price or quantity, the sessions took roughly 30 minutes
on average. Note that there are no practice periods, neither with nor without interaction
between subjects, and thus, no unobservable learning confounds occur. The matching
of subjects is constant throughout a session (fixed partner matching). In total, 240 students of economic fields participated in the experiment. Subjects were recruited via the
ORSEE platform (Greiner, 2015) and the hroot platform (Bock et al., 2014) and participated only in one of the treatments (between-subject design). The protocol for each
session is identical to the protocol described in Subsection 3.3.2. The total length of
a session from subjects’ entering to leaving the lab was about one hour. The average
payoff per subject was EUR 17.70.
6.2.3 Results
The experimental data amounts to 12 Bertrand and Cournot duopolies and triopolies,
each, as well as 14 (16) Bertrand (Cournot) quadropolies. Before analyzing the exper-
170
6.2 Experiment with symmetric firms
TABLE 6.7: Average degrees of tacit collusion across treatments.
Treatment
N
j Nash
jWalras
Friedman
B2
12
0.832
(0.249)
0.916
(0.124)
0.750
B3
12
0.605
(0.324)
0.737
(0.216)
0.556
B4
16
0.436
(0.267)
0.577
(0.200)
0.438
C2
12
0.627
(0.550)
0.907
(0.138)
0.936
C3
12
0.397
(0.484)
0.759
(0.193)
0.831
C4
14
0.206
0.603
(0.374)
(0.187)
Standard deviations in parentheses.
0.750
F IGURE 6.1: Average degrees of tacit collusion j Nash over periods across treatments.
Cournot Treatments
-.4 -.2 0 .2 .4 .6 .8 1
Bertrand Treatments
0
10
20
30
40
Duopoly
50
60
0
Triopoly
10
20
30
40
50
60
Quadropoly
imental data longitudinally, Table 6.7 and Figure 6.1 provide an overview of experimental data12 based on the level of independent cohorts over all 60 periods.13 Similar
to the previous meta-study, the Friedman index predicts the degree of tacit collusion
poorly in terms of j Nash (r = 0.049, p = 0.670), but is significantly correlated with jWalras
(r = 0.409, p = 0.001).
12 Note
that one duopoly in treatment C2 is exceptionally competitive. In particular, its average degree of
tacit collusion based on Nash profits lies almost three standard deviations below the treatment mean.
All results reported in the following hold if this outlier is dropped.
13 Here, collusion degrees over all 60 periods as measured relative to the Nash equilibrium are displayed.
Appendix B.3 depicts an analogous figure for collusion degrees measured relative to the Walrasian
equilibrium.
171
Chapter 6 Number Effects and Tacit Collusion in Oligopolies
For an in-depth analysis of firms’ longitudinal behavior, a mixed-effects model is ran
to control for different base levels of tacit collusion in cohorts via a random intercept as
well as for different time dependencies due to learning via a random slope. Thus, the
estimated model is
E
jk,t
= b0 + x k
+ b Duopoly · Duopoly
+ b Quadropoly · Quadropoly
+ b Cournot · Cournot
+ ( b Period + b Period,k ) · t
+ ek,t ,
E as the average degree of tacit collusion of all firms’ prices or quantities in
with jk,t
cohort k in period t. Table 6.8 shows the estimated coefficients for both degrees of tacit
collusion.14 All results reported in the following with respect to prices or quantities
hold also if the degree of tacit collusion is measured by transaction prices, i.e., prices
weighted by the quantities sold.
R ESULT 6.2. In the experiment with symmetric firms, the degree of tacit collusion based on the
Nash or the Walrasian equilibrium is significantly higher in markets with two firms than in
markets with three as well as four firms, and significantly higher in markets with three firms
than in markets with four firms, everything else being equal.
In line with the meta-analysis, the duopolies show, on average, a statistically significant 20 pp higher degree of tacit collusion than triopolies based on Nash predictions.
Moreover, and in contrast to the meta-analysis, quadropolies show a statistically significant 19 pp lower degree of tacit collusion than triopolies. The similar effect sizes of
both coefficients indicate not only a strictly monotonic, but a linear number effect in the
degree of tacit collusion. According to a Wald test, the equality of the absolute value
14 Note
that due to the dualism of the competition model used in the experiment, the degrees of tacit
collusion measured by prices or quantities coincide.
172
6.2 Experiment with symmetric firms
TABLE 6.8: Multilevel mixed-effects linear regressions of tacit collusion on number of competitors and competition model under competition between symmetric firms.
Baseline: Symmetric Bertrand Triopoly (B3).
Covariate
(1)
j Nash
(2)
jWalras
Duopoly
0.204⇤⇤
(0.092)
0.144⇤⇤⇤
(0.045)
Quadropoly
0.193⇤⇤
(0.087)
0.167⇤⇤⇤
(0.043)
Cournot
0.211⇤⇤⇤
(0.072)
0.008
(0.036)
Period
0.001
(0.001)
Constant
0.663⇤⇤⇤
(0.075)
Cohorts
Observations
78
4,680
0.001⇤
(< 0.001)
0.776⇤⇤⇤
(0.037)
78
4,680
Standard errors in parentheses.
⇤ p < 0.10, ⇤⇤ p < 0.05, ⇤⇤⇤ p < 0.01
of treatment dummy coefficients cannot be rejected (c2 (1) = 0.00, p = 0.946). Measured
relative to the Nash equilibrium, Bertrand competition colludes more than Cournot
competition. In fact, the increase of 21 pp in the degree of tacit collusion is similar to
the effect size found in the meta-study and also to the effect of an additional competitor
in a market as measured above. Thus, with regard to number effects, the experiment
replicates the findings of the meta-analysis with respect to duopolies and triopolies, but
also identifies a significant effect between triopolies and quadropolies.
With respect to the Walrasian-based degree of tacit collusion measure the data shows a
significant 14 pp (17 pp) increase (decrease) in duopolies (quadropolies) compared to
triopolies. Hence, there is also a monotonically decreasing, approximately linear trend
of the degree of tacit collusion as the number of firms in the market increases. These
findings indicate that subjects do indeed react to differences in theoretical predictions.
Moreover, there is a small yet significant negative time trend in the degree of tacit collusion if measured relative to the Walrasian Equilibrium.
173
Chapter 6 Number Effects and Tacit Collusion in Oligopolies
In summary, the results attained from the experiment with symmetric firms provide
support for the original conjecture of a strictly monotonic relationship between the
number of competitors and the degree of tacit collusion. In fact, the measured effect
sizes point to a linear trend with regard to the degree of tacit collusion relative to the
Nash equilibrium.
6.3 Experiment with asymmetric firms
As tacit collusion has been attributed to be driven by symmetry of firms in the economic
literature (Mason et al., 1992; Ivaldi et al., 2003), number effects in oligopolies may also
interact with the (a)symmetry of firms. Therefore, two additional experimental treatments with asymmetric, i.e., vertically differentiated, firms are considered. Thereby, the
experiment focusses on the comparison of triopolies and quadropolies, which clearly is
the most interesting case in the light of the previous results.
Consideration of asymmetric firms also has a high practical relevance, particularly in
the context of network industries, e.g., telecommunications and energy, which are still
characterized by a dominance of the former (state-owned) monopolist. Therefore, a
market structure with one incumbent and two or three entrants competing against each
other is conceived in the following. In particular, this market structure resembles the
regularities of many European mobile telecommunications markets, which are comprised of one large dominating and two or three smaller network operators, totaling at
three or four cellular networks in each national market.
6.3.1 Experimental design and procedures
For means of comparability to the previous experimental treatments, the experimental design introduced in Subsection 6.2.1 is extended to allow for asymmetric firms.
Asymmetry is implemented by establishing a single firm (i.e., the incumbent) with a
174
6.3 Experiment with asymmetric firms
higher quality good than the remaining—two or three—firms (i.e., the entrants). As
consumers value quality, the incumbent’s market share is higher than that of an entrant
for identical prices. Equivalently, for equal market shares the incumbent may charge a
higher price for its good than the entrants. In this vein, wk constitutes the reservation
price of firm k’s consumers in the model and thus, may be interpreted as the quality
of firm k’s product. Consequently, if the quality of one firm’s product is higher than
that of the other firms, i.e., wk > w
k,
the former has higher market power that results
in a higher equilibrium price, market share, and thus, profit. The extent of asymmetry
in product quality can be expressed by a single parameter, W = w Incumbent
wEntrant ,
which denotes the markup quality of the incumbent’s good compared to the entrants’
goods. See Appendix A.1 for the analysis of the model with horizontal as well as vertical differentiation.
Two additional asymmetry treatments—an asymmetric Bertrand triopoly (B3A) and
an asymmetric Bertrand quadropoly (B4A)—are considered. The parametrization for
entrants is the same as for firms in the symmetry treatments, i.e., wEntrant = 100, l = 1,
and q = 23 . Motivated by common market shares in European telecommunications markets, W however is now greater than zero and chosen such that the incumbent’s Nash
equilibrium profit is 50% higher than an entrant’s Nash equilibrium profit. Thus, the
incumbent’s market share with regard to its proportion of joint Nash equilibrium profits is 3/7 ⇡ 43% in a triopoly and 1/3 ⇡ 33% in a quadropoly. As market power is a
relative rather than an absolute concept, holding the relative profit markup of the incumbent constant has two important advantages over alternative approaches, such as
holding the incumbent’s market share constant. First, this allows to normalize entrants’
equilibrium profits such that they are the same as in the symmetry treatments (see below), which increases comparability across the symmetry and asymmetry treatments.
Second, the additional relative market power of the incumbent compared to any single
entrant is independent of the number of firms which increases comparability between
asymmetric triopolies and quadropolies. For the two asymmetric Bertrand treatments
a Nash equilibrium profit markup for the incumbent of 50% corresponds to W = 6.10
175
Chapter 6 Number Effects and Tacit Collusion in Oligopolies
in triopolies and W = 4.79 in quadropolies. The theoretical predictions of the one-shot
game for both asymmetry treatments are listed in Table A.2 in Appendix A.1.
In order to further ensure comparability, the same scaling and normalization as in the
previous experiment is applied: First, the action space of the incumbent is scaled such
that the JPM prices of all firms coincide at a price of 50 in an action space of [0, 100].
Second, profits are standardized such that an entrant would have the same Nash equilibrium gains as a firm in any of the symmetry treatments. Consequently, incentives
to deviate from the theoretical Nash prediction are equal for entrants and symmetric
firms in the previous experiment. The same scaling factor is applied to the entrants’
profits as well as to the incumbent’s profits so that the asymmetry in market power is
not affected.
Except for an additional paragraph in the experimental instructions explaining how
one of the firms differs from the others, the exact same experimental procedures are
followed for the asymmetry treatments as previously for the symmetry treatments.15
Again, the experiment was run at the Karlsruhe Institute of Technology in Karlsruhe,
Germany, and participants were recruited via the ORSEE platform for sessions between
June and August 2015 and via the hroot platform for sessions in May 2016. None of the
104 students of economic fields participating in one of the two asymmetry treatments
had previously participated in one of the symmetry treatments. The participants’ payoff averaged at EUR 19.82.
6.3.2 Results
Similar to the symmetry treatments, there are 12 independent asymmetric Bertrand triopolies and 17 independent asymmetric Bertrand quadropolies. Summary statistics for
both new treatments are provided in Table 6.9. Means over cohorts are computed by averaging over all firms, i.e., the incumbent and each entrant are weighted equally. With
15 As
an example, the experimental instructions for the asymmetric Bertrand quadropoly treatment together with a screenshot of the experimental software are provided in Appendix C.3.
176
6.3 Experiment with asymmetric firms
TABLE 6.9: Average degrees of tacit collusion across asymmetry treatments.
Treatment
N
j Nash
jWalras
B3A
12
0.332
(0.296)
0.554
(0.197)
0.792
0.444
B4A
17
0.217
0.412
(0.148) (0.111)
Standard deviations in parentheses.
0.619
0.379
Friedman
Incumbent Entrant
regard to the incumbent, the Friedman index is not significantly correlated with j Nash
(r = 0.160, p = 0.408), but significantly correlated with jWalras (r = 0.330, p = 0.081).
Regarding the entrants, the Friedman index—in line with the symmetry treatments—
predicts j Nash rather poorly (r = 0.295, p = 0.120) but has some explanatory power for
jWalras (r = 0.453, p = 0.014). Furthermore, the Friedman index predicts that the incumbent faces a higher incentive to tacitly collude—or, vice versa, that it has a lower
incentive to deviate from a collusive agreement. Thus, according to the Friedman index,
the entrants are supposed to be the drivers of competition. In the experimental sample, the incumbent indeed chooses, on average over both treatments, about 4 pp more
collusive prices than the entrants if measured relative to the Walrasian equilibrium.
This difference is significant according to a matched-samples Wilcoxon signed-rank test
(z = 2.87, p = 0.004) and coincides with the disparity between the Nash predictions for
the incumbent and entrants (see Table A.3 in Appendix A.1). In contrast, prices in terms
of the degree of tacit collusion relative to the Nash equilibrium vary neither largely nor
significantly between firms.
For an analysis of firms’ behavior in the asymmetry treatments, a similar mixed-effects
model as for the symmetry treatments is employed to control for different base levels
of tacit collusion in cohorts via a random intercept as well as for different time depen-
177
Chapter 6 Number Effects and Tacit Collusion in Oligopolies
dencies due to learning via a random slope. Thus, the estimated model is specified
as
E
jk,t
= b0 + x k
+ b Quadropoly · Quadropoly
+ ( b Period + b Period,k ) · t
+ ek,t ,
E as the average degree of tacit collusion of the incumbent’s and the entrants’
with jk,t
prices in cohort k in period t. Table 6.10 provides estimated coefficients for both measures of the degree of tacit collusion.
R ESULT 6.3. In the experiment with asymmetric firms, the degree of tacit collusion based on
the Nash or the Walrasian equilibrium is significantly higher in markets with three firms than
in markets with four firms, everything else being equal.
The Nash-based degree of tacit collusion is, on average, 21 pp higher in triopolies than
in quadropolies. In line with the previous findings from the symmetry treatments, this
difference is statistically significant and similar with respect to the effect size. Also consistent with the results under symmetry, the Walrasian-based degree of tacit collusion is
significantly higher in markets with three firms than markets with four firms. Furthermore, a negative time trend of prices due to an end-game effect can be found for both
measures. These results hold if only the entrants’ degree of tacit collusion is used.16
In summary, the empirical results under asymmetry replicate the findings under symmetry and thus support the general conjecture of a strictly monotonic relationship.
Whereas the relative effect on tacit collusion with regard to an additional competitor
is similar across symmetric and asymmetric market structures, the absolute degree of
tacit collusion for a market with a specific number of firms may differ, as a comparison
of Tables 6.7 and 6.9 indicates.
16 If
only the incumbent’s degree of tacit collusion is considered the difference between triopolies and
quadropolies is statistically insignificant according to the specified mixed-effects model.
178
6.3 Experiment with asymmetric firms
TABLE 6.10: Multilevel mixed-effects linear regressions of tacit collusion on number of competitors under Bertrand competition between asymmetric firms.
Baseline: Asymmetric Bertrand Triopoly (B3A).
(1)
j Nash
(2)
jWalras
Quadropoly
0.213⇤
(0.120)
0.203⇤⇤
(0.081)
Period
0.004⇤⇤⇤
(0.001)
0.002⇤⇤⇤
(0.001)
Constant
0.497⇤⇤⇤
(0.092)
0.665⇤⇤⇤
(0.062)
Covariate
Cohorts
Observations
29
1,740
29
1,740
Standard errors in parentheses.
⇤ p < 0.10, ⇤⇤ p < 0.05, ⇤⇤⇤ p < 0.01
While the detailed comparison of symmetry and asymmetry treatments is relegated to
Appendix B.3, it is briefly highlighted that the experimental results indicate support
for previous theoretical and empirical studies in their findings that symmetry seems
to facilitate tacit collusion (see, e.g., Mason et al., 1992; Ivaldi et al., 2003; Fonseca and
Normann, 2008). In particular with regard to quadropolies, asymmetry is a significant
driver of competition, with the degree of tacit collusion relative to the Nash (Walrasian)
equilibrium being 18 pp (13 pp) lower in asymmetric than symmetric markets with
four firms. Put into context, the effect size of implementing asymmetry by increasing
the market power of a single firm is comparable to the number effect on tacit collusion
between markets with two and three firms as well as between markets with three and
four firms.
In an effort to rule out social preferences—which may be viewed as an artifact of a laboratory setting—as a dominant motive for subjects’ decisions to collude less under asymmetry, subjects’ social value orientation is measured in an ex post questionnaire based
on Murphy et al. (2011). A comparison of the social value orientation index reveals no
differences between incumbents and entrants or subjects in triopolies and quadropolies
(see Appendix B.3). In conclusion, these findings suggest that social orientations cannot
explain why asymmetric firms collude less than symmetric firms.
179
Chapter 6 Number Effects and Tacit Collusion in Oligopolies
6.4 Discussion
The question that motivates this chapter is rather blunt: How many competitors are enough
to ensure competition? Evidently, it would be utterly unscientific to propose an answer
to this question disregarding the particular characteristics and circumstances in a given
market. But even if “case-by-case analysis implies that there is no ‘magic number’”
(Vande Walle and Wambach, 2014, p.10), the findings reported here point to systematic
effects with regard to tacit collusion that should be given careful consideration by competition and regulatory authorities when assessing the question of how to achieve and
safeguard effective competition in a market.
To this end, the three presented studies provide comprehensive evidence based on either existing or new oligopoly experiments, considering a different number of firms
(two vs. three vs. four firms), different modes of competition (price vs. quantity competition) and different degrees of market power (symmetric vs. asymmetric). First, the
meta-study on the extant literature studying number effects in experimental oligopolies
provides robust empirical support that tacit collusion is significantly higher in markets
with two firms compared to markets with three firms as well as to markets with four
firms. In contrast, neither intra-study nor inter-study evidence confirms a significant
effect between markets with three firms and markets with four firms. Thus, the extant
literature does not provide any evidence for the original conjecture of a strictly monotonic relationship between the number of competitors and the degree of tacit collusion
under Bertrand or Cournot competition.
However, there are several limitations to the meta-analysis itself as well as the individual studies that compare markets with three and four firms. First, the data collected
from experimental studies for the meta-study is heterogeneous in quality, i.e., it is either obtained from the authors directly, from tables in the article, or even retrieved
from figures. As a consequence, the granularity of the data varies across studies. Data
on the level of independent observations from sessions is only provided for half of the
studies considered here and hence, intra-study treatment differences are not replica-
180
6.4 Discussion
ble nor testable for the remaining studies. Second, the number of experimental studies surveyed in the meta-analysis is rather low, especially with regard to effects between triopolies and quadropolies, and thus the results are based on a small number
of observations. Moreover, within the individual studies, the number of independent
observations—in particular the number of independent quadropoly observations—is
also rather low. In consequence, the meta-study as well as individual studies may simply lack the statistical power to detect a potential number effect between triopolies and
quadropolies (cf. List et al., 2011; Bellemare et al., 2014).
Therefore, two experiments are conducted that further test the relative competitiveness
in triopolies and quadropolies based on a dataset with a considerably larger number
of independent observations and an experimental design that exploits the duality between differentiated Bertrand and Cournot competition. As a result, a significant effect
between markets with three firms and markets with four firms is found for both symmetric and asymmetric distributions of market power. Remarkably, the effect size of a
20 pp lower degree of tacit collusion is very similar from four to three firms as well as
from three to two firms across symmetric and asymmetric treatments. This points not
only to a strictly monotonic trend, but to a linear relationship between the number of
firms and the degree of tacit collusion measured relative to the Nash equilibrium. Figure 6.2 summarizes these findings and depicts the number effect in prices/quantities
across symmetric Bertrand and Cournot treatments.
Furthermore, the results both confirm and shed new light on previous insights. First,
the results indicate that already the judgment of the competitiveness of a certain mode
of competition (price vs. quantity competition) depends on the point of reference
(Nash vs. Walrasian equilibrium). In particular, Suetens and Potters (2007) suggest
that Bertrand colludes more than Cournot. However, as the empirical analysis reveals,
this holds only with respect to Nash equilibria. Instead, if tacit collusion is measured
with regard to Walrasian equilibrium, the opposite may hold. This finding of the metastudy is in line with the stronger competition predicted by the Nash equilibrium under
Bertrand competition than under Cournot competition. Second, the findings are in line
181
Chapter 6 Number Effects and Tacit Collusion in Oligopolies
F IGURE 6.2: Number effects across symmetric treatments.
40
Cournot Treatments
60
Bertrand Treatments
50
Quantity
70
60
80
90
0
100
10
20
Price
30
40
50
- JPM -
Duopoly
Triopoly
Quadropoly
Treatment average
Duopoly
Triopoly
Quadropoly
Nash equilibrium
with the notion that tacit collusion is more likely to emerge among symmetric rather
than asymmetric firms.
Whereas the conducted experiments address the limitations of the meta-study, they
are at the same time limited to the specific parametrization used. This applies to both
the specific demand parameters as well as to the way in which prices and profits are
normalized and scaled across treatments in order to maximize comparability. Also the
asymmetry between competitors may be parametrized in various ways, e.g., based on
differences in Nash profits or based on absolute differences in product quality. Although there is considerable variation in parametrization across the studies included in
the meta-analysis, it cannot be ruled out that the results of the conducted experimental
studies are to some degree affected by the specific parametrization used.
As a further limitation of all studies it is noted that competition in experimental
Bertrand and Cournot oligopolies is merely considered with exogenously symmetric
or asymmetric firms but not in the context of endogenous merger formation. Further-
182
6.4 Discussion
more, neither the experiments considered in the meta-analysis nor the conducted experiments allow for investments in order to increase the market size, which arguably plays
an important role in most industries that are characterized by an oligopolistic market
structure. With regard to a risk-averse regulator not only averages of market outcomes
may be of interest. Instead, authorities may also take into account the effect, which
the number of firms and regulatory institutions have on the variance of the expected
competitive intensity, in order to minimize the possibility of tacit collusion. However,
based on the samples in the meta-analysis as well as both oligopoly experiments, there
are no significant differences in this regard.
These limitations give rise to future research on number effects in oligopolies. In particular, the variety of asymmetric market settings offers opportunities for scenario-specific
investigations of competitive effects. For instance, a decrease in the number of competitors in a market, e.g., through a merger, is also likely to affect the horizontal and vertical
differentiation of firms’ products. In this vein, a merger may introduce asymmetry in
the market power of the remaining firms and thus not only relax competition due to
the decrease in the number of firms but also foster competition. Therefore, further experimental studies in the spirit of current replication efforts (see, e.g., Camerer et al.,
2016) are desired to test the robustness and generality of the presented findings (see
Chapter 8 for a more elaborate discussion on this issue).
183
Chapter 7
Voluntary Open Access in Digital Services
Markets
T
HE Internet as an online platform ecosystem now encompasses a diverse set of
web pages, mobile apps and services, shopping and product comparison sites, as
well as content and media platforms. The rapid expansion of this ecosystem and its user
base is spurred by the entry of new platforms and increased specialization of individual
outlets. At the same time, a small number of general-purpose platforms has emerged,
especially social networks. These platforms are now often characterized as the gateways
to the Internet (Arakali, 2015; Barnett, 2010) as they are used by the vast majority of
Internet users and often serve as a starting point for their online activities. The most
prominent example, Facebook, served 1.59 billion monthly active users worldwide at
the end of 2015 (Facebook, 2016) and a 74% share of US online adults in 2014 (Pew
Research Center, 2016). Next to the sheer amount of user data, these platforms—and
Facebook in particular—are in the possession of unique personal data as its social and
personalized services naturally require accurate and extensive information input from
its users.
Expansion and specialization at the edges on the one hand, together with increased
concentration at the core on the other hand, have created a situation in today’s online
This chapter is based on joint work with Jan Krämer and Michael Wohlfarth (Krämer et al., 2016).
185
Chapter 7 Voluntary Open Access in Digital Services Markets
platform ecosystem where user data and usage data are collected and stored by different
entities. In consequence, there emerges a rationale for data sharing between online
content providers (CPs)—and even competitors—to combine data sources in order to
enhance knowledge about users and potential customers. Targeting of consumers based
on a superior data set is viewed as a major competitive advantage in these markets,
particularly in the context of online advertising. Next to purchasing usage data ex post
in the form of collected tracking data through specialized intermediaries or markets
(cf. Montes et al., 2015), CPs may try to directly access and share users’ data across
platform boundaries. So-called social logins (Gafni and Nissim, 2014; Janrain, 2014),
which allow users to authenticate with third-party CPs through their social network
account have been established as one of the most popular instruments to share user
and usage information among online outlets. The most popular social login1 , Log in with
Facebook2 , e.g., allows websites and mobile apps to access users’ public profile including
demographic data, email address and friends. Moreover, access to extended profile
properties such as the history of users’ likes or recorded web activities may be requested
from the user.3 In return, Facebook obtains comprehensive data on users’ activities at
the third-party outlet through the programming interfaces of the login service. In fact,
Facebook states that it “can analyze [a third-party’s] app, website, content, and data for
any purpose, including commercial” in its Platform Policy.4
Whereas ex post data sale and purchase transactions are frequently conducted without
the knowledge of the data subject concerned, consumers choose the social login service
actively and voluntarily. Such user control has been found to be vital for successful
marketing campaigns in the context of social media (Fournier and Avery, 2011; Tucker,
2014). In fact, the widespread popularity of social logins indicates that reduced transaction costs as well as enhanced personalization and customization provide real added
value to users (see Acquisti and Varian, 2005, for a related scenario where personalization alters consumers decisions in the context of price discrimination). Already in 2010,
1 According
to LoginRadius (2015) 97% of websites that offer a social login employ Facebook’s login.
https://developers.facebook.com/docs/facebook-login. Accessed March 14, 2016.
3 For a comprehensive list of accessible properties see https://developers.facebook.com/docs/
facebook-login/permissions. Accessed March 14, 2016.
4 See https://developers.facebook.com/policy. Accessed March 14, 2016.
2 See
186
two years after the launch of the service, Facebook stated that 250 million consumers
and two million third-party websites used its social login and that the installed base
was growing by 10,000 sites per day (Grove, 2010). The current installed base is estimated at almost 10 million web sites (SimilarTech, 2016). According to Janrain (2014)
51% (88%) of consumers have used (encountered) a social login at least once and more
than half of the users (64%) “are more likely to return to a website that remembers them
without a username and password”. In this vein, increased user engagement and interaction are often quoted as main reasons for outlets’ decision to adopt a social login (see,
e.g., Gigya, 2015). However, the emergence of social logins as universal authentication
and authorization services has not been without criticism. Next to privacy and security
concerns and users’ inability to internalize potential long-term disadvantages (Breuer
et al., 2015; Kontaxis et al., 2012), strategic threats due to dependency on and exploitation by a dominant undertaking (Feingold, 2013) as well as legal risks (Van Der Sype
and Seigneur, 2014) have been cited as major issues for CPs who decide to adopt a
social login.
Despite the practical relevance, online CPs’ incentives to offer direct access to user information or to share usage data by means of a social login together with the ensuing
effects on competition have not yet been addressed in depth by the academic literature. This chapter scrutinizes outlets’ strategic decisions whether to offer and/or adopt
a social login, identifies the feasible market outcomes and highlights the implications
for the involved decision makers. The incentives and fundamental trade-offs that drive
adoption decisions and impact the profitability of CPs are examined based on a stylized model of horizontal competition between special-intererst CPs (who may adopt
the login), and advertising competition between special-interest CPs and the generalinterest CP (who may offer the login). Whereas there are situations where all parties are
mutually better off, it is demonstrated that special-interest CPs may implement a social
login even in cases where this decision ultimately makes them worse off, i.e., they may
find themselves in a prisoner’s dilemma-like situation. On the contrary, there may also
arise situations where the general-interest CP does not offer the social login although
it would be beneficial to consumers. This may provide a rationale for potential policy
187
Chapter 7 Voluntary Open Access in Digital Services Markets
interventions in the form of Open Access based on no-discrimination considerations,
which have traditionally been debated at the network layer (see Chapter 2), but may
be extended to the platform layer as discussed by Graef et al. (2015) and Easley et al.
(2015). This could in consequence restrict services provider’s discretion with regard to
discriminatory practices.
The remainder of this chapter is structured as follows: Section 7.1 surveys related literature strands on single sign-on systems, access to user data and user privacy, online advertising and targeting, which inform the model framework presented in Section 7.2. The model is analyzed and the feasible market outcomes are presented in
Section 7.3. Section 7.4 discusses policy questions together with possible extensions
and limitations.
7.1 Related literature
Although social login mechanisms have so far been explicitly addressed either by scholars concerned with usability of services and user acceptance (Egelman, 2013; Gafni and
Nissim, 2014), from a technical perspective (Breuer et al., 2015; Kontaxis et al., 2012), or
from a legal point of view (Van Der Sype and Seigneur, 2014), there are several related
strands of literature in the domains of economics, management and marketing, as well
as information systems, which will be discussed in the following.
One strand of studies explores the factors that affect consumers’ decision to use a social login rather than a website’s own registration service. For example, Kontaxis et al.
(2012, p.321) point to an additional “social dimension to the browsing experience” due
to users’ ability to share, rate and interact with content. These features require the
sharing of user and usage data between the respective social network and the content
provider. Based on an exploratory survey, Gafni and Nissim (2014) identify familiarity
and convenience as factors that positively affect users’ readiness to opt for a social login
in a world in which Internet users face an increasing number of websites that require
188
7.1 Related literature
authentication. In such cases, social logins avoid the need for multiple (different) username and password combinations, evade repetitive registration processes and minimize the effort to update and maintain accurate information in the case authentication
properties change. In a laboratory experiment, Egelman (2013) examines convenience
benefits of the Facebook Login relative to perceived privacy costs and finds that for the
majority of subjects the former outweighs the latter even if this requires the disclosure
of extensive personal information and although subjects are well-aware of the scope of
collected data.
The technical literature views social login services (Ko et al., 2010) in the history of (enterprise) single sign-on systems (SSOs), i.e., systems that allow for centralized and federated identity management across remote and distributed resources (Pashalidis and
Mitchell, 2003). Next to the design and implementation of authentication protocol standards, e.g., OAuth (Sun and Beznosov, 2012; Chen et al., 2014), studies have been concerned with the security of data transmission (Wang et al., 2012) and users’ control over
their personal data in different services contexts (Dey and Weis, 2010; Kontaxis et al.,
2012). In essence, these studies suggest that social logins reduce users’ transaction costs
in a distributed (web) context, if technical systems are designed and implemented securely and with regard to users’ privacy concerns. However, Sun et al. (2010) also note
that many SSOs, in particular open source systems, failed due to a lack of adoption
incentives for involved platforms and content providers.
Additionally, social logins introduce new business opportunities and thus adoption
incentives through the access to external data sources and potential sharing agreements among competitors. With regard to such voluntary access relationships among
competitors, well-known economic trade-offs for the involved transaction parties arise
(Boudreau, 2010; Mantena and Saha, 2012; West, 2003). In general, a firm needs to consider whether additional access revenues (e.g., wholesale revenues in vertical markets,
cf. Ordover and Shaffer, 2007) outweigh the business stealing effect (Mankiw, 1986),
due to competitors’ access to the firm’s exclusive resources. Moreover, wholesale access has been found to soften competition in downstream markets (Bourreau et al.,
189
Chapter 7 Voluntary Open Access in Digital Services Markets
2011), whereas with regard to price discrimination a common data pool may intensify
competition (Fudenberg et al., 2000; Montes et al., 2015). Thus the interplay between
data sharing and competitive intensity is a priori unclear.
Incentives to collect data were scrutinized with regard to firms’ ability to price discriminate based on historic usage data when consumers react to such practices (see
Fudenberg and Villas-Boas, 2007, 2012, for summaries of behavior-based price discrimination in digital markets). More specifically, Acquisti and Varian (2005) show that
personalization of services which increases consumers’ valuation over time of use may
be critical to make price discrimination profitable if consumers act strategically. Access
to users’ personal data may invoke a range of privacy issues that have been examined
by a growing strand of theoretical and empirical literature (see Acquisti et al., 2016, for
an extensive survey). With regard to social logins, authors have criticized inter alia the
loss of anonymity, revelation of social information, loss of traceability in cases of a data
breach, propagation of advertisements, and disclosure of user credentials as potential
threats to users’ privacy (Kontaxis et al., 2012). However, the fact that users deliberately
decide to use the social login opposed to alternative registration options suggests that
the positive effect on users’ valuation outweighs potential privacy concerns (Egelman,
2013).
Over and beyond the ability to offer personalized services and prices, data collection
and analysis is driven by the desire to improve the display of online advertising and to
better match advertising to users’ preferences by the means of targeting. Evans (2008)
provides an overview of online advertising markets and characterizes key features
that differentiate them from traditional media markets (Anderson and Coate, 2005).
Bergemann and Bonatti (2011) show that on the one hand targeting benefits consumers
through improved matching between consumers’ preferences and advertised products,
but on the other hand, targeting increases market concentration in the advertising industry.
Online advertising allows for a much more diverse set of pricing models and performance metrics than for offline advertising (Asdemir et al., 2012): Consumers’ response
190
7.1 Related literature
to online advertising may either be measured based on click-throughs, i.e., the immediate engagement with the advertisement, or based on view-throughs, i.e., the more comprehensive long-term effect on consumers’ purchase decision (Bleier and Eisenbeiss,
2015). Next to potential brand building effects (Yoo, 2009) and long-term impact on
consumers’ decision (Drèze and Hussherr, 2003; Manchanda et al., 2006; Lewis and
Reiley, 2014), which are neglected by click-based or action-based measures, the academic literature has identified moral hazard as a potential impediment to the adoption
of pricing models associated with pure click-through measures (Animesh et al., 2010;
Asdemir et al., 2012; Liu and Viswanathan, 2014) or action-based measures (Hu et al.,
2015). View-through (which includes click-through) is therefore commonly considered
as the relevant performance measure in display advertising markets (Hamman and
Plomion, 2013) and especially by advertisers in social networks. According to market
research (Ross, 2015) more than 60% of total advertising budget spent at Facebook is allocated via optimized CPM (oCPM), a metric where advertisers are billed on the basis of
impressions, while Facebook optimizes bidding for the effective cost per impression according to advertisers’ preferred mix of user interaction (actions, reach, clicks or social
impression).5 With close to 24% of total revenues in 2014 (eMarketer, 2015) and a third
of total display impressions to US Internet users in 2011 (comScore, 2011), Facebook
has become the clear market leader in display advertising and its share is estimated to
grow further over the coming years (Marshall, 2015).
Through the social login, Facebook and other social networks may be able to drastically
improve their ability to display targeted advertising and thus to increase view-through.
According to Iyer et al. (2005) ubiquitous data gathering allows firms to gain “much better information on consumers, their preferences and their media habits” (p.461). Moreover, the advent of new media in combination with fragmentation of traditional media
allows firms to differentiate and delineate between target segments. In particular, Iyer
et al. show that targeting increases advertisers profits by reducing wasted impressions
to consumers who are not interested in an advertiser’s product. In economic theory,
5 Further information is provided by Facebook’s documentation of pricing models and advertising objec-
tives: https://www.facebook.com/business/help/355670007911605. Accessed March 14,
2016.
191
Chapter 7 Voluntary Open Access in Digital Services Markets
consumers benefit from improved targeting of advertising, because the displayed ads
are perceived as more relevant to their interest and therefore also more informative
(Butters, 1977; Anand and Shachar, 2009). This in turn leads to increased user engagement for respective advertisers. Although survey-based research also identifies potential adverse effects due to better informed advertising (Turow et al., 2009), empirical
studies generally support the hypothesis that data collection and analysis increases the
effectiveness of displayed advertisements by the means of targeting (Braun and Moe,
2013; Goldfarb and Tucker, 2011a,b; Urban et al., 2014) and personalization (Ansari
and Mela, 2003; Bleier and Eisenbeiss, 2015; Tucker, 2014).6 In particular, Goldfarb and
Tucker (2011a) find that targeted ads, which, e.g., would match the content of the visited website, exhibit a significantly higher effectiveness than conventional non-targeted
ads. Moreover, in a study of advertising effectiveness on Facebook, Tucker (2014) finds
that personalization of ad impressions in addition to targeting further increases the likelihood that users click on displayed ads if consumers maintain (perceived) control over
their privacy. With regard to retargeting effectiveness, Lambrecht and Tucker (2013)
and Bleier and Eisenbeiss (2015) show that advertisers need to be accurately informed
about the current status of consumers’ purchasing decision process. This further emphasizes the necessity for a firm’s global view on consumers’ activities over and beyond
its own platform.
In the following, a game-theoretic model is developed that offers a microfoundation
and characterization on how social logins affect the competition between CPs for users,
as well as the competition between CPs and the social network for advertising revenues. Research gaps of the extant literature are addressed by explicitly considering the
effects of data access and information sharing among competitors on CP’s ability to attract users as well as its ability to generate advertising revenues. Moreover, factors that
impact CPs decision to offer access to user data and/or usage data are examined. More
specifically, it is shown that a higher competitive intensity between special-interest CPs
6 See Tucker (2012) for a discussion of adverse effects on advertising effectiveness if user perceive data col-
lection as excessive. However, these issues may be resolved by (perceived) user control over disclosure
of their personal data as shown by Tucker (2014).
192
7.2 A model of coopetition in online advertising markets
makes information sharing with a general-interest CP more likely even if this is ultimately detrimental to the special-interest CPs.
7.2 A model of coopetition in online advertising markets
Before providing the details of the model, the basic competitive setting is outlined.
7.2.1 Competitive setting
The considered online advertising ecosystem includes two competing special-interest
content providers (CP s, s 2 { A, B}), one general-interest content provider (CP G) and
one advertiser (Z). The special-interest CPs may be thought of as specialized web
sites that offer a narrow range of content in the same domain of interest (e.g., celebrity
news or financial news), whereas the general-interest CP offers a much broader range
of complementary content (e.g., a general news site). In the context of this chapter
and the preceding motivation it will be convenient to think of CP G as a social network provider (e.g., Facebook, Twitter, LinkedIn or GooglePlus), although the insights
of the model are not restricted to this scenario. The important aspect of the model is
that the special-interest CPs operate in the same domain and are thus competing directly for users, whereas the general-interest CP is already used by all Internet users
under consideration. This captures the relevant competitive dynamics of the Internet,
where, on the one hand, users split their attention between a general-interest site (e.g.,
Facebook) and a particular special-interest site, but, on the other hand, choose to pay
attention to only one of the two comparable special-interest sites (e.g., they visit either
hellomagazine.com or ok.co.uk, but not both). In other words, whereas users multihome between CP s and CP G, they single-home between CP A and CP B.
193
Chapter 7 Voluntary Open Access in Digital Services Markets
Furthermore, it is assumed that all CPs offer their content free of charge to Internet users
and derive revenues by charging a price, p, for showing display advertisements on their
sites. To date, this is the prevalent business model on the Internet (cf. Dou, 2004; Evans,
2009; Anderson, 2012) and consequently, this is the dominant modelling assumption
in the related literature (see, e.g., Athey et al. (2014); Choi and Kim (2010); Kourandi
et al. (2015)).7 To fix ideas, it is assumed that CPs offer advertisement space based
on pricing per impression (i.e., they adopt a cost-per-mille (CPM) model). Although,
there exist also other means to sell advertisement space, in particular based on pricing
per click (cost-per-click (CPC) model) or per transaction on the advertiser’s site (costper-action (CPA) model), CPM is still widely adopted by large CPs (e.g., Facebook,
Google Ad Sense) and consequently, this is the standard assumption used in the context
of display advertisements markets (see, e.g., Anderson and De Palma (2013); Johnson
(2013); Reisinger (2012)).8 In order to focus on the competition between CPs, it is further
assumed that there is a single advertiser, Z, that wishes to buy advertisement space for
a special-interest ad (in the same domain as the special-interest CPs) at any one of the
CPs. As will be described in detail subsequently, CPs differ in their targeting ability
and, due to competition and split attention between CPs, each CP reaches a different
subset of the users at any given point in time. Depending on a CP’s ad pricing relative
to how many view-throughs can be achieved at this CP, the advertiser will choose at
which subset of the CPs to advertise in order to maximize its profit. Thus, it is important
to see that, although the general-interest CP is not in direct competition for users with
any of the special-interest CPs, it is still in competition with them for users’ attention to
7 However,
this assumption is mainly made for convenience and clarity of the model, because it would
only be required that CPs derive a significant portion of their total revenues from advertising. Other
revenue streams (e.g., from transaction fees or subscriptions) are not considered here, because they
are not directly relevant to the analysis of the competitive dynamics of the online advertising market,
which is the focus of this analysis.
8 Moreover, as will be seen later, the CPM model allows for the formulation of an intuitive microfoundation of how advertisement prices are formed and affected by the competition between CPs. The
corresponding microfoundation would be more complicated and hence, less intuitive under a CPC
model, without providing additional insights. The results remain qualitatively unaffected as long as a
CP’s advertisement price increases with the attractiveness of that CP to users (which is especially the
case for enhanced targeting, see Athey and Gans, 2010). For a more detailed analysis of the trade off
between the CPC and CPM model, see Asdemir et al. (2012).
194
7.2 A model of coopetition in online advertising markets
ad impressions. This basic competitive set-up of the model is summarized in Figure 7.1.
F IGURE 7.1: The competitive setting of the model: Each user chooses exactly one of the two
special-interest CPs (single-homing) and splits attention between the specialinterest CP and the general-interest CP (multi-homing). All CPs are ad-financed
and receive advertisement revenues according to their relative ability to reach adrelevant users.
CP G
advertiser
CP A
CP B
xor
registration
Competition for users
7.2.2 Details of the model
In the following, a more detailed description of the building blocks of the model, the
strategic variables of each of the entities involved, as well as the timing of actions will
be provided.
Internet users
There is a unit mass of heterogeneous Internet users that have a natural
preference for one of the two special-interest CPs. Users’ preference for the CPs is denoted by x and assumed to be uniformly distributed between zero and one (Hotelling,
1929). The two special-interest CPs are horizontally differentiated and located at either
end of the users’ preference spectrum, i.e., CP A at x = 0 and CP B at x = 1. Thus, a
type x consumer derives utility of U A ( x ) = u A
tx or UB ( x ) = u B
t (1
x ), when he
consumes the content of CP A or CP B, respectively. Thereby, us denotes the base utility
195
Chapter 7 Voluntary Open Access in Digital Services Markets
that is derived from the viewing experience and usability of the site (e.g., the quality of
the content or the hassle of the login procedure), and t is the degree of competition between the two special-interest CPs. When t is large, the users’ innate preference for the
CPs becomes more important, such that competition on the basis of us becomes weaker.
Users will visit only the special-interest CP which gives him the highest utility. The respective user demand for CP s is denoted by Ds . Furthermore, it is assumed that us is
large enough, such that the market is fully covered, i.e., at any time D A + DB = 1.
Besides one of the special-interest CPs, the users will also visit the general-interest CP.
In order to model how users split attention between the two considered CPs, it is assumed that there are two time periods, indexed by t = 1, 2. In each time period, a user
visits the special-interest CP with probability d and the general-interest CP with probability 1
d. Thus, d and 1
d are referred to as the screen attention probability of CP
s and CP G, respectively. Consequently, in each time period t, a special-interest CP s
expects to be viewed by a total of Ds,t = Ds d users, whereas the general-interest CP G
expects a total of DG,t = ( D A,t + DB,t ) (1
d) = 1
d viewers per time period.
Content providers All content providers k 2 { A, B, G } receive revenues from selling
advertisement space to the advertiser. In principle, each CP demands a price for displaying the ad on its site in both time periods. In order to focus on the relevant aspects
of the model the costs of providing content of each CP are normalized to zero and it
is further assumed that advertisements can be displayed at zero marginal costs. Moreover, it is important to highlight that the special-interest CP s is only in competition
with CP G for views, because users single-home between special-interest CPs. Thus,
prices will depend on the submass of consumers, Ds , s = A, B that is considered. Therefore, CP s’s price is denoted by ps and CP G’s price for the submarket Ds is denoted by
p G (s) .
196
7.2 A model of coopetition in online advertising markets
Consequently, the profit of CP s is Ps = ps , whereas the profit of CP G is PG =
Âs= A,B pG(s) . While a CP seeks to maximize its profit from advertising, p will depend
on the CPs’ ability to reach ad-relevant users, which is influenced by two factors: (i)
the number of a CP’s viewers in time period t, which again depends on competition
and screen attention, and (ii) its ad targeting rate ak  1. A CP’s targeting rate denotes
which fraction of the viewers that actually see the ad belong to the advertiser’s target
group. More specifically, the targeting rate describes view-through, i.e., how well a CP
can transform views into relevant advertising impressions.
In this context, it should be emphasized that the model is set up to analyze the economic
effects that arise in competition for the advertising budget of a particular advertiser
targeting a specific audience. Thus, attention is restricted to the subset of the online advertisement ecosystem that is relevant for this advertiser, comprised of special-interest
websites (targeted at roughly the same audience as the advertiser) and general-interest
websites. Evidently the special-interest CPs are considered by a much smaller number
of advertisers (which is approximated by a single advertiser here) than the generalinterest CP, who will therefore display a larger variety of ads. The revenue streams
that may arise for the general-interest CP from these other advertisers are not explicitly
modeled, because they are not relevant for the economic effects that arise in any given
advertisement submarket, which is considered here.
CP G may choose to offer CP s a social login. If CP s chooses to adopt the social login
feature on its site, the two CPs implicitly agree to cooperate by means of sharing information about their users. Formally this has two implications:
First, it is assumed that CP s’s base utility, us increases by q
0. This can be motivated
by a better user experience at CP s, e.g., because the same credentials can be used at
both CPs, which limits password fatigue and lowers transaction costs of registration, or
because information sharing enables better personalization of content and better integration of the services of both CPs. Of course, there may also be countervailing effects
197
Chapter 7 Voluntary Open Access in Digital Services Markets
on the users’ experience, in particular due to concerns of privacy or a single point of
failure.9 However, it is assumed that the overall users’ experience is better with the
social login. More formally:
us =
8
>
>
<ubs ,
if CP s does not use the social login
>
>
:uls = ubs + q
if CP s uses the social login,
whereby the superscript l denotes the case where the login is used and the superscript
b denotes the base case without the social login. Furthermore, it is assumed that CPs
are symmetric in the base case, i.e., ubA = ubB = ub .
Second, due to information sharing about the user, the targeting rates of CP G and CP
s are increased by a factor of fG
ak =
8
>
>
<ab ,
k
1 and fs
1, respectively. Consequently,
for the mass of users, Ds , of CP s that does not use the social login
>
>
:al = min{fk ab , 1},
k
k
for the mass of users, Ds , of CP s that uses the social login.
Again, it is assumed that special-interest CPs are symmetric in the base case, i.e., abA =
abB = abs .
Advertiser The advertiser wants to place an informative ad that is targeted at a specific target audience. The goal of the advertiser is to generate attention for its product
or service, which, e.g., can be a visit to its online- or offline-store some time after the ad
has been viewed. The effectiveness of an ad is thus measured with respect to the rate
of effective view-through (Hamman and Plomion, 2013), which includes the case of a
click-trough, but also considers lagged responses of consumers (Asdemir et al., 2012;
Bleier and Eisenbeiss, 2015). In particular, it is assumed that the value of the first ad impression that is displayed to a user belonging to the target audience is v = 1. Moreover,
9 For
example, the "Log in with Facebook" feature was unavailable for several hours in September 2015,
preventing users to login at all CPs that used this social login (Burlacu, 2015).
198
7.2 A model of coopetition in online advertising markets
because the ad is informative, it is assumed that all subsequent ad impressions on the
same user are wasted and thus, do not create additional value for the advertiser (for a
similar assumption see, e.g. D’Annunzio and Russo, 2015; Calvano and Jullien, 2012;
Athey et al., 2014; Ambrus et al., 2016).10 The objective of the advertiser is to select the
subset of CPs at which to advertise in order to maximize its profit, i.e.,
P Z = G A (n A
p A ) + G B (n B
pB ) +
Â
s= A,B
G G (s) ( n G (s)
p G ( s ) ),
where Gs (GG(s) ) is an indicator function, which returns one if an ad is placed for the
mass of Ds users at CP s (CP G) and zero otherwise. Moreover, ns (nG(s) ) denotes the
expected number of ad-relevant viewers from the total mass of users Ds at CP s (CP G)
that see the ad for the first time and thus have a value of v = 1 for the advertiser. In
order to keep the presentation concise n will be referred to simply as “view-throughs"
in the following.
Structure and timing The following four-stage game is considered:
Stage 1: The general-interest CP G decides whether to offer a social login for the
special-interest CPs s.
Stage 2: Special-interest CPs s simultaneously but independently decide whether to
adopt the social login and users decide which CP to use.
Stage 3: All CPs simultaneously set advertisement prices pk .
Stage 4: The advertiser decides at which CPs to advertise.
10 In
fact, it is only required that the marginal value of subsequent ads is decreasing (cf. Anderson and
De Palma, 2013). The assumption that subsequent views have zero marginal value is thus merely made
to keep the analysis simple. Empirical research on advertising effectiveness shows that additional
impressions of the same ad are indeed less valuable for the advertiser (see, e.g., the survey by Simon
and Arndt, 1980).
199
Chapter 7 Voluntary Open Access in Digital Services Markets
7.3 Competitive effects of the social login and market outcomes
The subgame perfect equilibrium is determined by solving through backward induction, i.e., beginning in Stage 4 and proceeding backwards.
Stage 4: Advertiser’s decision In order to decide at which CPs to advertise, Z calculates the expected view-throughs per CP, nk . Since the special-interest CP s is only in
competition with CP G for views, it suffices to consider only the mass of users at CP
s, i.e., Ds . In the first period every ad impression will be the first for a visitor, so that
CP s expects view-throughs of ns,1 = as dDs . In reverse, the remaining Ds
dDs users
have visited CP G in the first period, and thus CP G expects nG(s),1 = aG(s) ( Ds
dDs )
view-throughs in the first period from the mass of users that multi-home between CP
s and CP G. Recall that CP G’s targeting rate can differ between different masses of
users, Ds , because it depends on whether or not CP s has adopted the social login.
This is denoted by aG(s) . In the second period, only those users are relevant for the
advertiser that have not seen the ad in the first period at any of the two CPs, i.e.,
Ds
ns,1
nG(s),1 . Note that in period two, the mass of users Ds again redistributes
its attention randomly between CP s and CP G according to d, i.e., users that have been
targeted at CP s may now visit CP G, and vice versa. Thus, CP s and CP G expect
to generate ns,2 = as d( Ds
200
ns,1
nG(s),1 ) and nG(s),2 = aG(s) (1
d)( Ds
ns,1
nG(s),1 )
7.3 Competitive effects of the social login and market outcomes
new view-throughs in the second period, respectively. In summary, this yields a total
number of view-throughs by CP k of11
(7.1)
ns = as dDs + as d( Ds
| {z }
|
period t = 1
(7.2)
nG =
Â
as dDs
a G ( s ) ( Ds
{z
period t = 2
aG(s) ( Ds dDs ) + aG(s) (1
{z
} |
s= A,B |
d)( Ds
period t = 1
dDs ))
}
as dDs
{z
a G ( s ) ( Ds
period t = 2
dDs )) .
}
It is important to see that nk does not only depend on CP k’s own targeting rate, ak , but
also on the other CP’s targeting rate. This is the basis for competition between the CPs
in the advertising market. With regard to the mass of users Ds , the advertiser faces the
decision to display advertising either exclusively at CP s or CP G, or at both outlets at
the same time. Obviously, the latter option maximizes view-throughs n = nG(s) + ns .
However, the advertiser maximizes profit Gs (ns
ps ) + G G (s) ( n G (s)
pG(s) ) and may
decide to switch exclusively to a CP if it can gain a higher net benefit through a lower
price p for the view-throughs. For example, the advertiser will decide to reach the mass
of Ds consumers exclusively through CP s if nes
ps > Âm=G(s),s (nm
pm ). Thereby
nes , which denotes the total view-throughs at s when CP s is the advertiser’s exclusive
outlet, is larger than CP s’s total view-through when CP s is not exclusive. This is
because CP s can generate more view-throughs in the second period, as users have
not been able to view the advertisement already at the other CP in the first period.
The advertiser’s decision where to advertise will thus depend on view-throughs, as
identified above, and CPs’ prices, which are determined next.
Stage 3: CPs’ ad pricing First, see that exclusive advertising cannot be an equilibrium
outcome, because CPs do not have marginal costs of advertising and thus the excluded
CP, which currently experiences zero profits from advertising, would always be better
11 Note
that it has been implicitly assumed that the advertiser has chosen to display the ad at all CPs,
which is indeed the equilibrium outcome. A complete characterization of the model would also require
to specify nk in those cases in which the advertiser choose only a proper subset of the CPs. It is easy to
see that nk can then be derived by setting ak = 0 for all CPs that are not chosen by the advertiser.
201
Chapter 7 Voluntary Open Access in Digital Services Markets
off by lowering its advertising price, such that it may be considered again by advertiser Z. This competition for selection by Z constrains the CPs’ advertising prices. In
equilibrium prices must therefore be chosen such that Z is indifferent between selecting
both CPs and selecting one CP exclusively. Hence, equilibrium prices can be derived
by solving the following system of equations
(7.3)
n G (s)
p G (s) + ns
ps = neG(s)
(7.4)
n G (s)
p G (s) + ns
ps = nes
p G (s)
ps ,
which yields equilibrium prices of p⇤k ( Ds , as , aG(s) ). Note that equilibrium prices (and
thus CPs’ equilibrium profits) behave intuitively with respect to changes in the targeting rate. For example, in line with Athey and Gans (2010), a higher targeting rate,
which in turn yields more view-throughs, allows CPs to demand a higher equilibrium
price p⇤ and thus yields higher profits, i.e.,
∂pk⇤
∂ak
> 0. Moreover, competition between
CPs yields a negative effect of the rival’s targeting ability on a CP’s equilibrium profit,
i.e.,
∂ps⇤
∂aG(s)
< 0, and
⇤
∂pG
(s)
∂as
< 0, respectively.
Stage 2: Special-interest CPs’ social login adoption Provided the social login is offered by CP G, each CP s decides independently whether to adopt the social login by
comparing anticipated advertising profits P⇤k given targeting rates ak and market share
Ds . More specifically, each CP s trades off the positive effect on consumers’ valuation for their services and/or a potentially improved advertising effectiveness against
the negative effect of intensified competition in the advertising market. On the one
hand, the adoption of the social login increases the special-interest CP’s base utility by
q, which may generate a competitive advantage over the other special-interest CP and
expand its market share from Dsb to Dsl . However, if the rival CP also adopts the social
login, then both CPs offer the same base utility again and thus, Dsb = Dsl . Moreover,
the social login may improve the special-interest CP’s targeting rate based on the ad-
202
7.3 Competitive effects of the social login and market outcomes
ditional information over its users and therefore increases its advertising effectiveness,
i.e., als = fs abs , with fs
1. On the other hand, data sharing may also improve the
general-interest CP’s targeting rate, i.e., alG(s) = fG abG(s) , with fG
1, which negatively
affects the special-interest CP’s advertising revenues, as shown above. In summary,
four different scenarios can be distinguished: Either both special-interest CPs adopt
(scenario l, l) or do not adopt (scenario b, b) the social login, or only one CP adopts the
social login (scenarios l, b and b, l). The resulting normal form game is summarized in
Table 7.1 and the corresponding profits are derived in Appendix A.4.
TABLE 7.1: Normal form game representing special-interest CPs’ social login adoption decision.
CP B
Social Login (l) No Login (b)
CP A
Social Login (l)
No Login (b)
l,l
(Pl,l
A ,P B )
l,b
(Pb,l
A ,P B )
b,l
(Pl,b
A ,P B )
b,b
(Pb,b
A ,P B )
In the following, it is shown that if special-interest CPs are symmetric, either both or
none adopt the social login. Thereby, CP s considers the net effect Dl,d
b,d
:= Pl,d
s
Pb,d
s
of the social login on its anticipated profits, given its rival’s decision d 2 {b, l }. CP s
adopts the social login, given that its rival does not adopt it, if and only if Dl,b
Pl,b
s
b,b
:=
Pb,b
s > 0, which is satisfied if its increase in targeting rate is above the critical
threshold
fs
(7.5)
⇣
⌘
1
b
·
a
(
d
1
)
(
t
+
q
)
f
+
t
+
q
G
G (s)
d abs (t + q )
⇣
[ (fG abG(s) (d 1))2 + 2abG(s) (d 1)(fG dabs ) + (dabs
!
⌘1/2
+ q (t + q )(1 + fG abG(s) (d 1))2 ] (t + q )
.
1)2 t
203
Chapter 7 Voluntary Open Access in Digital Services Markets
On the other hand, CP s adopts the social login, given that its rival also adopts it, if
and only if Dl,l
b,l
:= Pl,l
s
Pb,l
s > 0, which is satisfied if its base-targeting rate does not
exceed the critical threshold
⇣
1
·
abG(s) (d
d abs t
⇣
[t (fG abG(s) (d
fs
+ (d + abs )2 (t
(7.6)
q)
1 ) t fG + t
⌘
1))2 + 2(d
1)(abG(s) (tfG d(t
!
⌘1/2
q )abs + t )]t
.
2d(t
q )abs )
As special-interest CPs are symmetric and decide simultaneously, it follows directly
that both CPs will adopt the login (universal adoption) whenever Dl,b
b,b
> 0 and no
CP will adopt the login otherwise.12 Due to the symmetry of the special-interest CPs,
in both cases Ds = 1/2.
In situations where both CPs A and B have an unilateral incentive to adopt the login,
i.e., Dl,b
b,b
> 0, resulting profits under universal adoption may not necessarily exceed
profits in the initial setting without the login. Instead, special-interest CPs may find
themselves in a prisoner’s dilemma-like situation, where relative competitive benefits
due to the social login cancel out if the other CP adopts the social login as well. In
consequence, increased advertising competition by the general-interest CP G due to
information sharing via the social login may outweigh any positive impact on CP s’s
own targeting rate. This is the case if Dl,l
b,b
:= Pl,l
s
Pb,b
s < 0, which is satisfied if and
only if
(7.7)
1
fbs <
·
d abs
12 Formally
204
⇣
⇣
abG(s) (d
(fG abG(s) (d
1 ) fG + 1
⌘
1))2 + 2(d
it can be shown that fs > fs .
1)abG(s) (fG
dabs ) + (dabs
1)2
⌘1/2
!
.
7.3 Competitive effects of the social login and market outcomes
Stage 1: General-interest CP’s social login offer
Anticipating special-interest CPs’
adoption decision and ensuing effects on advertising prices, the general-interest CP decides whether or not to offer the social login. To this end CP G weighs the positive effect
on its own targeting ability against the negative effect of the rival’s improved targeting
ability, both of which will affect competition in the advertising market. In general, CP
G is willing to offer the social login if Ul,l
b,b
:= Pl,l
G
Pb,b
G > 0, since the symmetric
special-interest CPs always make the same adoption decision. This condition is satisfied if
fes <
(7.8)
2
abG(s) (fG
1) ( d
1) + 2 (dabG(s) + fG
2 d fG abs
1)
.
Possible market outcomes: Based on the previous analysis and the therein derived
thresholds, the feasible market outcomes that may arise can now be fully characterized. As special-interest CPs are symmetric, it will generally suffice to consider any
submarket Ds to discuss the possible market outcomes. Thereby, it is most insightful to
delineate the different market outcomes in terms of CP s’s and CP G’s increase in targeting rate due to the social login. In particular, the market outcomes are determined
by the critical thresholds i) fs , ii) fbs and iii) fes derived above, that denote i) when the
special-interest CPs adopt the social login, ii) when special-interest CPs are actually
better off by adopting the social login, and iii) when the general-interest CP offers the
social login.
In total there are six possible market outcomes, which are illustrated in Figure 7.2.
I:
The social login is offered and adopted. All CPs are better off, i.e., profits
with the social login are higher than without it.
II:
The social login is offered and adopted. Whereas the general-interest CP is
better off, special-interest CPs are worse off (prisoner’s dilemma-like outcome).
205
Chapter 7 Voluntary Open Access in Digital Services Markets
III:
The social login is not offered, but special-interest CPs would be willing to
adopt it. However, adoption would make them worse off.
IV:
The social login is offered, but special-interest CPs do not adopt it.
V:
The social login is not offered, but special-interest CPs would be willing to
adopt it. Adoption would make them better off.
VI:
The social login is neither offered nor would it be adopted by special interest
CPs. All CPs would be worse off with the social login.
Evidently, the social login is either not offered, or not adopted, or both in market outcomes I I I - V I. Furthermore, it is immediately obvious that the social login is neither
always offered nor always adopted when it would be socially optimal. First, the social
login is always beneficial to consumers, who experience an increase in base utility by q,
everything else being equal. Thus, the market outcomes I I I to V I are inefficient with
respect to consumers surplus, because the social login is not offered or not adopted
here. Moreover, in market outcome I I special-interest CPs are worse off despite the
fact that they have voluntarily adopted the social login. Finally, in market outcome V
special-interest CPs would be better off by using the social login, but are not offered
this option. Consequently, in five of the six possible market outcomes a market failure
may occur, whereas only in outcome I the market outcome is always efficient.
R ESULT 7.1. In market outcome I I (voluntary) adoption of the social login leaves the specialinterest CPs worse off. Generally, provision of the social login may create market failures in
market outcomes I I to V I, whereas it is always efficient in market outcome I.
Comparative statics Subsequently, the conditions under which the social login is offered and adopted are further examined. Therefore, it is investigated how the critical
thresholds fs , fes and fbs change ceteris paribus in response to a change in one of the
model’s exogenous parameters. While the details of the comparative statics are rele-
206
7.3 Competitive effects of the social login and market outcomes
F IGURE 7.2: Illustration of possible market outcomes: The social login is offered in outcomes
I, I I and IV and not offered otherwise. The social login is adopted in outcomes
I and I I and not adopted in outcome IV. In outcome I special-interest CPs are
better off and in outcome I I they are worse off by adopting the social login. Note:
The figure is derived for abs = 0.5, abG(s) = 0.5, t = 0.5, q = 0.1, d = 0.5. Not all
market outcomes may exist for all feasible parameter constellations.
!"
#"
!
$"
!
III
VI
V
!s
II
I
IV
!G
207
Chapter 7 Voluntary Open Access in Digital Services Markets
gated to Appendix A.4, it can be concluded that the model yields quite intuitive results. An increase in the CPs’ targeting rate without the login, as for CP s (aG(s) for CP
G), makes it less likely for CP s (CP G) to adopt (offer) the social login. A similar effect
can be observed for an increase in a CP’s screen attention, d for CP s ((1
d) for CP G),
which translates into the same effect as if one’s own targeting rate is increased while
simultaneously the rival’s targeting rate is decreased. Overall an increase in one’s own
screen attention therefore decreases incentives for the adoption (for CP s) and the offer
(for CP G) of the social login.
The parameters aG(s) , as and d, which affect the competition in the advertising market,
impact all three critical thresholds. By contrast the parameters q and t, which affect
the competition for users between special-interest CPs, impact only threshold fs . Intuitively, as users’ benefit of the social login increases (increase in q) or as the competition
for users becomes weaker (decrease in t), the social login becomes less relevant for the
competition for users and thus, it is adopted less likely by the special-interest CPs.
R ESULT 7.2. As a special-interest (general-interest) CP’s targeting rate or screen attention
increases, it becomes less attractive to adopt (offer) the social login. The more special-interest
CPs are in competition for users, the more the social login is adopted.
Illustrative market scenarios To conclude the analysis two specific market scenarios
are highlighted that are illustrative in the sense that they represent extrema of the feasible spectrum of possibilities.
First, consider the case where the special-interest CPs have already attained a high targeting rate and thus only the general-interest CP will be able to increase its targeting
rate due to information sharing via the social login, i.e., fG > fs ⌘ 1. In this case the
special-interest CP cannot gain a competitive advantage in the advertising market from
adopting the social login. It will base its adoption decision therefore purely on the ex-
208
7.4 Discussion
pected impact of the social login on the competition for users. In particular, if q > 0
the prisoner’s dilemma-like situation will prevail, as special-interest CPs are symmetric and none can commit not to use the social login. Eventually both will adopt the
login and compete again head to head, each attaining 50 percent market share. In the
end, only the general-interest CP, and, of course, the consumers, benefit from the social login. From Figure 7.2 it is easy to see that at fs = 1 only two market outcomes
are feasible. The general-interest CP will always offer the social login and either both
special-interest CPs adopt it and are worse off (outcome I I), or they do not adopt it
(outcome IV).
Second, consider the polar case where the social login does not offer any net benefit
to consumers, i.e., q = 0. In this case the special-interest CPs will base their decision
whether or not to adopt the social login purely on the effect in the advertising market.
This means that fs and fbs coincide in this case, because special-interest CPs will only
adopt the social login if and only if it is eventually profitable for them. Thus, market
outcomes I I and I I I do not exist and the prisoner’s dilemma-like situation does not
arise here.
7.4 Discussion
Social logins are single sign-on solutions provided predominantly by social networking
sites that allow users to register with and login at otherwise unaffiliated CPs using their
existing login credentials of the social network. While a social login provides added
value to the website users, they also imply that user data is shared between the CP that
employs the social login and the social network that offers it. Whereas the former effect
is relevant for the competition for website visitors, the latter effect is relevant for the
competition in the advertising market, as data sharing improves the websites ability
to offer targeted advertising. The presented model captures these two dimensions of
209
Chapter 7 Voluntary Open Access in Digital Services Markets
competition and identifies the winners and losers with regard to the usage of social
logins in the various market outcomes that may arise. In particular, it is shown that the
adoption of the social login may yield a prisoner’s dilemma-like situation for the CPs.
The competition for users may induce them to adopt the social login, but eventually
they may stand to lose their competitive advantage over the social networking site
in the advertising market. Moreover, it is demonstrated that social logins are likely
to yield a market failure because the login is not offered in cases where it would be
beneficial for users and the advertiser.
Although the model considers only a particular submarket, it is evident from the analysis that the social network wishes to offer the login only to those CPs from which it
can profit relatively more (in terms of increase in its targeting rate) than the content
provider that employs the social login. Indeed, although Facebook generally allows
every content provider to employ the social login, it lays out in its platform policy that
it may “enforce against [a] [...] website if [Facebook] concludes that [it] violates [its]
terms or is negatively impacting the platform" and that in this case it "may or may not
notify in advance".13 Thus, this analysis also raises an important policy question that
is related to the debate on net neutrality (Krämer et al., 2013; Easley et al., 2015) and on
Open Access (see Chapter 2). The Open Access debate at the telecommunications infrastructure layer centers around the question to what extent the operator of an access
network is allowed to discriminate between retailers that rely on access to the network
infrastructure. Hence, the network operator as a gatekeeper controls the access to an
essential upstream resource. Similarly one can raise the question here whether Facebook, who is also an important gatekeeper with regard to its mostly unique personal
data resources, should be allowed to discriminate between different content providers
in providing access through its Facebook login. Given the analysis in this chapter, it
13 See
https://developers.facebook.com/policy, ”Things you should know", number 6. Accessed March 14, 2016.
210
7.4 Discussion
is demonstrated that market failures may occur and thus there potentially is room for
welfare improving policy interventions.
Finally, possible model extensions and limitations are highlighted. First, the presented
analysis starts from the premise that the adoption of the social login provides a net
benefit to consumers. However, the model could be extended to also incorporate a net
loss in users’ utility, e.g., due to privacy concerns. In this case, the prisoner’s dilemma
would be reversed in the sense that special-interest CPs may not adopt the social login
although it would be beneficial to them. Second, special-interest CPs are assumed to
be symmetric. Although this is the obvious starting point, a much richer set of market
outcomes may arise in case CPs are asymmetric, e.g., with respect to their targeting
rates ex-ante, or with respect to their increase in net utility or targeting rate ex-post.
The prisoner’s dilemma-like situation may also be mitigated if markets were not fully
covered or if a CP would enjoy some additional elastic demand from users that does not
come from the rival CP. Third, some of the model parameters that are currently treated
as independent and exogenous may in fact be correlated and partially endogenous. For
example, the screen attention may depend on the net benefit that a CP offers, or a CP’s
targeting rate may depend on its demand. This will likely amplify the already observed
effects. At last, the presented analysis considers a monopolistic login provider as well
as a monopolistic advertiser. Although this seems to represent a reasonable assumption
with respect to the presence of market power and network effects in online content and
digital services markets, it may be worthwhile to extend the analysis to competing login
providers and advertisers, respectively.
211
Chapter 8
Conclusion
T
HIS thesis studies the competitive and cooperative interactions in information
and communications technology markets, where firms require access to an es-
sential competitive resource. In the light of continuous technological innovation, firms
and policy makers alike are challenged by the consequences of transforming market
structures in the telecommunications industries and the emergence of new digital services markets. As outlined in this thesis, combining theoretical analysis and experimental evaluation provides a methodological foundation to examine market outcomes
under institutions that may govern these markets. By serving as a testbed for policy
proposals, this work aims at identifying the welfare implications of designated institutions and informing the design of new regulatory institutions prior to their implementation in the field. Specifically, the concept of Open Access is scrutinized with regard to
its implications for competition and its ultimate impact on consumers. The insights and
conclusions drawn from these investigations thus contribute to the scientific discourse,
but also support policy makers and practitioners when assessing the implications of
competitive and cooperative strategies in ICT markets.
213
Chapter 8 Conclusion
8.1 Summary and implications
In the following, the main findings of this thesis are summarized and discussed with
respect to their methodological and policy implications. The section is organized along
the research questions set out in Chapter 1.
Addressing Research Question 1, this thesis proposes a unified definition of Open
Access based on various notions offered by different stakeholders in the context of
telecommunications infrastructure markets. According to this understanding, Open
Access, in essence, mandates that access to an upstream resource is provided in a nondiscriminatory manner. In situations where the upstream resource is provided by a vertically integrated firm this requires non-discrimination between the integrated downstream subsidiary and an independent retailer as well as non-discrimination between
two independent retailers. Chapter 2 demonstrates that Open Access may thereby be
achieved by a range of alternative regulatory institutions. At a fundamental level, these
institutions may differ with regard to i) the vertical structure that is allowed or prescribed at the supply-side of the upstream resource, ii) the degree and form of publicsector participation, and iii) the level of access and the corresponding wholesale products.
In consequence, the choice for a specific regulatory institution involves several tradeoffs that must be recognized by authorities and policy makers. More generally, the surveyed economic literature points to a trade-off between static and dynamic efficiency
that each regulatory institution balances differently. Based on the effects identified in
the theoretical and empirical literature, Chapter 2 suggests a guideline to assist policy
makers and regulatory agencies in their decisions according to their primary objective.
In this vein, it is made explicit that regulatory policy needs to weigh and prioritize
objectives. To this end, traditional regulatory institutions such as price regulation and
vertical separation have been examined thoroughly by the extant literature. In contrast,
214
8.1 Summary and implications
the review points to research gaps with regard to alternative regulatory regimes which
may be better suited to balance the efficiency trade-off. Most notably, a regime based on
mandated non-discrimination may be able to safeguard competition by creating a level
playing field, but at the same time foster investment incentives as wholesale charges
are not subject to ex ante price regulation.
Margin squeeze regulation, in particular, has been proposed as an alternative institution to price regulation and referred to as a possible means to ensure Open Access in
telecommunications markets where infrastructure competition has been established.
Previous investigations of margin squeeze regulation, however, have predominantly
been concerned with the application in markets with a single dominant supplier of the
wholesale resource. Therefore, Research Question 2 aims specifically at a market structure, where a second integrated firm is able to self-supply the upstream good. Chapter 4
examines whether margin squeeze regulation could improve market outcomes relative
to no regulation in the case of a wholesale monopoly based on a game-theoretic model.
This market scenario resembles the current situation in many fixed telecommunications markets. Although margin squeeze regulation may prevent foreclosure and thus
could be beneficial for independent retailers, consumers are found to be unambiguously worse off due to higher retail prices. Remarkably, total welfare may increase in
some cases, due to higher producer surplus under margin squeeze regulation. These
findings contradict the rationale underlying the current legislative implementation in
Europe: competing infrastructures do not represent exogenous competitive constraints
as assumed, but are likely to react strategically to the implementation of new regulatory
rules. In consequence, rather that the wholesale price level falls, the overall retail price
level rises under margin squeeze regulation.
Empirical evidence from a laboratory experiment provides further support for the theoretical findings that margin squeeze regulation does not benefit consumers in a market characterized by infrastructure competition and a wholesale monopoly. Moreover,
215
Chapter 8 Conclusion
Chapter 5 extends the analysis of margin squeeze regulation to wholesale competition, i.e., a market scenario where both integrated firms may offer wholesale access to
their respective upstream resource. In this scenario, theory would predict that margin
squeeze regulation benefits consumers, as foreclosure is prevented and the competitive outcome remains as the single equilibrium prediction. However, experimental
evidence contradicts this prediction as consumer welfare does not increase under margin squeeze regulation compared to no regulation. In fact, empirical findings indicate
that under regulation retail market prices may increase to the detriment of consumers,
especially in cases where wholesale competition is effective in reducing prices below
the monopoly level. On the contrary, margin squeeze regulation may allow the retailer
to obtain higher profits due to lower access prices only in cases where a high degree
of tacit collusion among integrated firms impedes effective wholesale competition. As
findings on margin squeeze regulation contrast conventional wisdom and the intuitive
appeal of its justifying rationale, the general need for thorough theoretical and experimental evaluation of policy proposals is emphasized.
Infrastructure competition has often been articulated as the ultimate goal of sectorspecific regulation and may thus be viewed as a sufficient criterion to abstain from
price regulation. Furthermore, previous theoretical studies have pointed to wholesale
competition as an effective means to achieve Open Access for independent retailers at
the competitive wholesale price (cf. Ordover and Shaffer, 2007; Brito and Pereira, 2010;
Bourreau et al., 2011). These studies, however, do not take into account the possibility of tacit collusion, which is frequently observed in markets with only few integrated
competitors (Parker and Röller, 1997). Therefore, Chapter 5 investigates Research Question 3 and yields the following result: wholesale and retail prices are found to be higher
under wholesale competition than under a wholesale monopoly. The surprising finding
that consumers are in consequence worse off under competition is rationalized by an
investigation of collusion incentives in an infinitely repeated game context. It is shown
that the symmetric market structure under wholesale competition facilitates tacit col-
216
8.1 Summary and implications
lusion relative to the asymmetric market structure under a wholesale monopoly. In
the latter case, the vertically integrated firm that does not provide access, benefits relatively less from tacit collusion. Moreover, it has a stronger incentive to price more
aggressively in the retail market as it does not face any opportunity costs in the form
of foregone wholesale revenues. In conclusion, these results point to a more general
dilemma of the Open Access rationale: whereas non-discrimination may provide the
basis for competition on equal terms, symmetry may facilitate coordinated behavior
among competitors to the detriment of consumers. In this vein, tacit collusion may
constitute a primary concern above and beyond the exercise of market power. Therefore, it is of particular interest that a complementary price commitment rule—which
does not change the theoretical prediction—significantly decreases the empirically observed degree of tacit collusion and in consequence leads to significantly lower market
prices as well as higher consumer surplus.
The relevance of tacit collusion for competition in concentrated markets motivates Research Question 4, which asks for the relationship between the number of competitors
and the degree of tacit collusion in a market. A meta-analysis of experimental studies
that have varied the number of firms provides general support for the notion that markets with a larger number of firms are less prone to tacit collusion than markets with
a lower number. In particular, previous studies confirm a significant negative effect on
tacit collusion from four to two as well as from three to two firms. Suprisingly, the extant experimental literature does not provide empirical evidence for a significant effect
from four to three firms under either Bertrand or Cournot competition, and therefore
cannot substantiate the assumption of a strictly monotonic number effect on tacit collusion. Therefore, Chapter 6 conducts two experimental studies with symmetric and
asymmetric distributions of market power that systematically investigate number effects for two, three and four firms, while also controlling for the competition model.
Based on a considerably larger number of independent observations compared to the
surveyed studies, the experiments find evidence for a significant number effect on tacit
217
Chapter 8 Conclusion
collusion from four to three firms as well as from three to two firms. In fact, the empirically observed increase in the degree of tacit collusion relative to the Nash equilibrium
is almost identical from four to three as it is from three to two. Indeed, this suggests a
linear number effect on the degree of tacit collusion for highly concentrated oligopolies.
Furthermore, the experimental evidence contributes to the notion that asymmetry between firms impedes coordinated behavior and leads to lower degrees of tacit collusion
compared to a symmetric market structure. Notably, market outcomes remain significantly above the theoretical prediction even in markets with four firms, demonstrating
subjects’ robust ability to sustain tacit collusion—even under a finite time horizon.
From a methodological perspective it is important to recognize the relevant design dimensions that affect the degree of observed tacit collusion in economic laboratory experiments. In this spirit, Research Question 5 is concerned with the consequences of
conducting competitive interactions in continuous rather than discrete time. Therefore,
Section 3.3 classifies extant studies in non-discrete time and then conducts an experiment to compare tacit collusion under continuous and discrete time, while controlling
for the competition model and the number of firms. In contrast to previous studies,
which have found that continuous time facilitates cooperation, e.g., in a prisoner’s
dilemma context (Friedman and Oprea, 2012), the obtained empirical evidence suggests that tacit collusion in an oligopoly competition context is significantly lower with
a continuous framework than with a discrete framework. This finding bears important methodological implications. First, it may inform the design of future oligopoly
experiments with regard to the consequences of the chosen mode of interaction (continuous vs. discrete). Second, these findings may be taken into account when results
of oligopoly experiments in discrete time—which applies to the vast majority of the extant experimental literature—are generalized to a context that is rather characterized by
continuous time. This is the case whenever decision makers can decide on the timing
of an action endogenously and thus asynchronous interaction may occur. Finally, given
the contrasting findings in previous comparisons of continuous and discrete time, the
218
8.2 Limitations and outlook
results suggest that whether continuous time facilitates or impedes cooperation hinges
critically on the specific game context.
Above and beyond the telecommunications infrastructure layer, Research Question 6
addresses the implications of voluntary access relationships among competitors at the
digital services layer. To this end, Chapter 7 examines the competitive effects of social logins that allow for the sharing of user and usage data between online content
providers and social networks. It is shown that content providers may adopt the social login even if this strategic decision makes them ultimately worse off, i.e., they find
themselves in a prisoner’s dilemma-like situation. The rationale for adoption is based
on the relative advantage that a content provider can gain over its horizontal competitor by sharing information with the social network. However, this advantage is offset if
all content providers decide to adopt the social login. In doing so, the content providers
allow the social network—the provider of the social login—to access a superior data basis and to increase its revenues in the advertising markets through a higher targeting
rate. Moreover, the analysis shows that the social network has incentives to discriminate between content providers if it can offer the social login on a discretionary basis.
More specifically, it may not offer the social login if this would threaten its position in
the advertising market relative to the respective content provider. In conclusion, this
shows that voluntary access regimes do not necessarily ensure Open Access with regard to data resources. In consequence, consumers may be worse off. On the contrary,
voluntary access offers may be implemented by a dominant gatekeeper to protect its
position, albeit not through exclusion, but through exploitation.
8.2 Limitations and outlook
This thesis has several limitations with regard to its scope and its methodology that
point to promising avenues for future research. Whereas, limitations specific to each
219
Chapter 8 Conclusion
study are discussed in the respective chapters, this section is dedicated to a summary
of the overreaching issues.
Foremost, the studies in this thesis analyze the underlying issues from a purely static
perspective and do not explicitly consider effects on dynamic incentives. More specifically, the thesis focusses on the question which institutions might improve static market
outcomes compared to alternative institutions, in particular to the benchmark scenario
of no regulation. Thus, the implicit premise is that regulatory regimes which abstain
from price regulation are more likely to have a positive effect on dynamic incentives.
This reasoning is based on the trade-offs between static and dynamic objectives identified for regulatory institutions in telecommunications infrastructure markets in Chapter 2, and the growing empirical evidence on the chilling effects of wholesale price regulation (inter alia Briglauer et al., 2013; Bacache et al., 2014) on investment incentives.
Moreover, a trade-off between static and dynamic efficiency has long been recognized
in economic theory going back to Schumpeter (1942) and found to apply to a more general context than infrastructure markets and price regulation (Aghion et al., 2005). Yet,
there is an ongoing academic discussion about the more precise characterization of the
trade-off’s functional form (Sacco and Schmutzler, 2011; Hashmi, 2013) and potential
interaction effects with varying market characteristics (Gilbert, 2006). With regard to
the current developments in telecommunications markets as well as antitrust cases in
digital services markets, the necessity to understand the relationship between static and
dynamic effects in specific application scenarios is only expected to grow (Bauer, 2014;
Parker and Van Alstyne, 2014). In these antitrust cases, authorities and courts are regularly challenged to weigh the benefits of competitive markets in a static sense against
incentives for innovative activity. Therefore, additional theoretical, experimental, and
empirical work is required to inform those decision makers about the consequences
of alternative regimes with regard to specific application scenarios and to explore and
design institutions that can balance the general efficiency trade-off.
220
8.2 Limitations and outlook
From a methodological perspective this calls for further research on how laboratory
experiments can be employed to investigate and compare dynamic effects across institutions. The evaluation of investment decisions thereby presents several challenges
because path dependencies and experimental subjects’ unobserved strategic reasoning
considerably affect experimental outcomes. Most notably, experimental subjects need
to form expectations about future market interactions and associated payoffs when
faced with investment decisions. Yet, these expectations may not realize accordingly,
and in consequence, subjects may base their subsequent decisions on forward induction, i.e., behavior that rationalizes past decisions, rather than backward induction, which
is regularly assumed in the theory to be tested. Whereas this issue can be mitigated
through a more constrained experimental design, i.e., a higher degree of experimental
control, such measures may limit the “behavioral freedom” of experimental subjects
(Morgan, 2005). In consequence, this may prevent interesting effects in the interplay
between dynamic and static considerations to arise in the first place. Nevertheless,
studies that have examined investment decisions in an economic laboratory context
point to potential approaches that can alleviate the aforementioned issues and provide
a methodological basis for future inquires (see, inter alia, Darai et al., 2010; Sacco and
Schmutzler, 2011; Aghion et al., 2014; Krämer and Vogelsang, 2016). In particular, Sacco
and Schmutzler (2011) query subjects’ beliefs in the investment stage and are thus able
to relate actions in subsequent periods to those beliefs. Future experimental work may
therefore focus on identifying and capturing individuals’ strategic rationales at the micro level without unintentionally framing and interfering with those decisions.
Two general shortcomings of the experimental method apply to this thesis in particular.
First, recent methodological analyses and replication studies have raised general concerns about the robustness and generalizability of individual empirical studies (Ioannidis, 2005; Simmons et al., 2011). Most notably, life sciences (Prinz et al., 2011; Begley and
Ellis, 2012) and psychological science (Open Science Collaboration, 2015) have found
themselves in a “credibility crisis” in consequence of a “reproducability crisis” (Baker,
221
Chapter 8 Conclusion
2016). As a result, these disciplines have introduced additional mechanisms and increased efforts to improve the reliability and robustness of empirical findings (see, e.g.,
Nosek et al., 2015). On the one hand, pre-registration of study designs (Schulz et al.,
2010) is emphasized in order to avoid false positive findings and minimize flexibility
in data collection and analysis (Schwab et al., 2011; Simmons et al., 2011) as well as
to mitigate publication bias (Easterbrook et al., 1991). On the other hand, ex ante statistical power analysis is traditionally required for medical trials in order to minimize
false negative findings, i.e., erroneous findings of non-significance despite a true effect
(Lachin, 1981; Lenth, 2001; Button et al., 2013). Finally and foremost, large replication
programs have been initiated to corroborate robustness of findings by individual studies (e.g., Errington et al., 2014; Klein et al., 2014).
In experimental economics, similar efforts are desired to safeguard scientific credibility through replication efforts that assess robustness of findings and methodological
standards that minimize confirmation bias (Roth, 1994). A recent replication study of
laboratory experiments (Camerer et al., 2016) and methodological guidelines regarding statistical power analysis for economic experiments (List et al., 2011; Bellemare
et al., 2014) represent starting points for further activities in this regard. In the spirit
of Engel (2007) and Suetens and Potters (2007), empirical meta-analyses as conducted
in Chapter 6 can shed light on the generalizability of findings by individual studies. In
particular, the parametrization decisions made in this thesis may be subject to future
robustness tests. As the parametrization of variables is inherently required when models are used as experimental instruments (see Chapter 3), the ensuing effects of specific
parameter sets and their generalizability bear important practical and methodological
implications.
The second general concern regarding economic experiments that study firm behavior
in a laboratory context is the lack of external validity. Although shortcomings in this
regard are inherent in the experimental methodology—which aims at internal validity
222
8.2 Limitations and outlook
as the primary objective (Guala, 2005)—there are possible approaches to mitigate those
shortcomings. First, external validity may be tested in experimental studies that go
beyond pure student samples and instead include experts and practitioners within the
designated context of application. Chapter 5, for instance, conducts a validation study
with expert subjects from the telecommunications industry, which allows for a comparison of outcomes with the student relative to the practitioners sample. Note, however,
that the inclusion of expert subjects involves additional challenges that need to be considered in the design and implementation of such validation studies. Experts may introduce implicit additional considerations based on their everyday experience into the
laboratory that however are not considered in the experimental context or even the underlying theory (Ball and Cech, 1996). Additional questionnaires and comprehension
tests can help to make these implicit assumptions and expectations explicit. In turn,
this allows experimenters to reinforce experimental control ex ante or at least allow for
control ex post by considering peculiarities in the analysis of experimental outcomes.
A further step to address external validity above and beyond the laboratory context is
the implementation of experimental field studies (Einav and Levin, 2010). In particular,
electronic markets and digital services may allow for controlled variation of market
institutions directly in the designated application context (see the discussion in Levin,
2013). In regulatory practice, new institutions may be implemented first in pilot studies
in order to test theoretical and experimental predictions prior to large-scale application
in the field. On the downside, these pilot studies may introduce additional strategic
incentives for the involved decision makers that researchers and regulators need to be
aware of.
Finally, experimental studies in this thesis have provided several results that point to
empirical regularities, which, however, are not (yet) well understood from a theoretical
perspective. Based on the empirical results at the macro level, the development of microfounded theory is desired to gain a more precise understanding of how the observed
223
Chapter 8 Conclusion
market outcomes evolve. For instance, Section 3.3 points to a systematically lower degree of tacit collusion under continuous time which however is in contrast to the few
theoretical analyses that suggest more cooperative behavior. In this context, broadening
the scope of methodological approaches may be beneficial to either conduct explorative
studies which could provide further stylized facts to inform theory building or to test
early explanatory theory hypotheses (see Section 3.1). For these purposes, agent-based
simulation could serve as a natural extension to laboratory experiments and as a means
to create more comprehensive testbeds for institutions (Duffy, 2006). In this vein, future work could build on previous simulation studies that have examined individual
strategies in the context of tacit collusion (Midgley et al., 1997; Erev and Roth, 1998;
Waltman and Kaymak, 2008) as well as the design of markets and institutions (Bichler
et al., 2010; Block et al., 2010; Ketter et al., 2013).
Above and beyond the aforementioned limitations, the findings of this thesis point to
themes that have received only little attention in the extant literature, but are believed
to play a major role in ICT markets. In particular, it is shown that tacit collusion and
anti-competitive coordination between competitors have a profound impact on market
outcomes, above and beyond the exploitation of market power by a single dominant
firm. More specifically, it is found that implicit cooperation among competitors may
have important ramifications for competition in and across vertically related markets.
As ICT markets become more mature and concentrated at the services layer, and at
the same time firms become more integrated across the value chain, these issues are
likely to be magnified (see A.T. Kearney, 2016, for a summary of recent market developments). Whereas the recent literature has provided important insights into market
power and its consequences in these markets, analysis of behavioral effects and collusion incentives in these contexts are currently scarce (see Hossain et al., 2011, for a rare
experimental analysis in the context of two-sided markets). As a result, it is for example
well-known that market concentration cannot be equated with market power per se in
two-sided markets (Evans and Schmalensee, 2013), but whether such market concen-
224
8.2 Limitations and outlook
tration may facilitate coordinated behavior to the detriment of competitors, suppliers
or consumers is currently unknown and thus open for future research.
Strategic decisions with regard to competition and cooperation as well as the balance
of those activities will represent key challenges for firms across the ICT value chain
(Nalebuff and Brandenburger, 1997; Bengtsson and Kock, 2000, 2014). In particular,
firms will need to weigh short and long-term consequences of cooperative vertical relationships with respect to effects on horizontal competition in their respective markets.
Moreover, competition will regularly take place between ecosystems, i.e., integrated
organizations, or partnerships that exceed clearly delineated market boundaries and
vertical value chain layers. Therefore, the management of access relationships and the
optimal degree of openness are crucial strategic variables for firms’ success in these
markets (Eisenmann et al., 2008). As shown in this thesis and in line with previous
studies, short-term benefits of cooperation may incentivize partnerships that prove to
hurt some firms in the long-term (Mantovani and Ruiz-Aliseda, 2015). From a policy
perspective, this points to new forms of market failures with regard to Open Access
among competitors. In contrast to telecommunications infrastructure markets, where
exclusion was the traditional concern, here, negative consequences on competition may
result from voluntary or discriminatory access agreements. In the spirit of this thesis,
policy proposals should thus be informed by microfounded theory and be subject to
empirical examination in the designated context of application prior to implementation in the field.
225
Appendix A
Theoretical Analyses
A.1 Oligopoly competition
Let the relevant industry consist of n 2 N firms. Each firm produces one good and
goods between firms are differentiated. Considering the representative consumer’s
utility function suggested by Singh and Vives (1984) and extending the generalization
by Häckner (2000), inverse demand for firm k 2 {1, ..., n} is given by
p k = wk
lk qk
g  qj
j6=k
with wk , lk > 0, 8k 2 {1, ..., n} and the degree of substitutability g. If g < 0 goods are
complementary, if g = 0 goods are independent of one another, and if g > 0 they are
substitutes. wk may be interpreted as quality and thus, differences among firms as
vertical differentiation. With substitute goods, wk is also firm k’s reservation price. lk
is the elasticity of inverse demand of firm k’s good. For simplicity, assume that lk =
227
Appendix A Theoretical Analyses
l, 8k 2 {1, ..., n} and let q = gl . This bounds q  1 with goods being perfect substitutes
if q = 1. The inverse demand for firm k then transforms to
(A.1)
!
l qk + q  q j .
p k = wk
j6=k
Note that firms are vertically differentiated, i.e., asymmetric, and that symmetry requires wk = w, 8k 2 {1, ..., n}. To calculate the demand for firm k, summarize Equation
(A.1) over all n firms, which results in
n
n
k =1
k =1
 pk =  wk
using Ânk=1 Â j6=k q j = (n
n
l
n
 qk + q (n
1) Â q k
k =1
k =1
!
1) Ânk=1 qk . Solving this for Ânk=1 qk yields
n
 qk =
k =1
1
l (1 + q ( n
n
1)) kÂ
=1
( wk
p k ).
As a transformation of this equation, noting that
n
 qk = qk +  q j ,
k =1
(A.2)
j6=k
n
 wk = wk +  w j ,
k =1
j6=k
and using Equation (A.1), firm k’s demand for non-perfect substitutes (q < 1) is given
by
(A.3)
qk =
( wk
pk )(1 + q (n
l (1
2))
q )(1 + q (n
q  j6=k ( w j
pj )
1))
provided that the quantity is non-negative and with n as the number of firms with
non-negative demand. Otherwise, if qk < 0, firm k exits the market and its demand is
zero.
228
A.1 Oligopoly competition
With costs normalized to zero and q
Pk = pk qk with price pk (qk , q
quantity qk ( pk , p
k)
k)
= {q1 , ..., qn } \ qk , firm k’s profit is given by
k
as a function of quantities in Cournot competition and
as a function of prices in Bertrand competition. In the following
analysis of Walrasian, Nash, and collusive equilibrium prices, quantities, and profits,
subscripts are used to differentiate between Bertrand and Cournot competition.
In the Walrasian equilibrium, also referred to as competitive
Walrasian equilibrium
equilibrium, firms are assumed to have no market power and hence, are price-takers
with all prices at marginal cost. Therefore, the Walrasian equilibrium is identical under
Bertrand and Cournot competition. Setting Equation (A.1) to marginal cost, i.e., zero, it
can be transformed to
qk (q
k)
=
lq  j6=k q j
wk
l
.
Summing over all n firms gives
n
 qk =
Ânk=1 wk
lq (n
l
k =1
1) Ânk=1 qk
,
which, using the previous Equation together with Equation (A.2), yields the Walrasian
equilibrium
qWalras
=
k
(A.4)
w k (1 + q ( n
l (1
2))
q )(1 + q (n
q  j6=k w j
1))
,
pWalras
= 0,
k
PWalras
= 0.
k
Nash equilibrium In the Nash equilibrium under Cournot competition firm k maximizes Pk with respect to its quantity qk given the other firms’ quantities q
best response is given by
qk (q
k)
=
wk
k.
Firm k’s
lq  j6=k q j
2l
229
Appendix A Theoretical Analyses
and its sum over all n firms amounts to
n
 qk =
Ânk=1 wk
k =1
lq (n
2l
1) Ânk=1 qk
.
Using the previous Equation together with Equation (A.2), the Cournot Nash equilibrium can be retrieved as
Nash
qCournot,k
=
Nash
pCournot,k
=
(A.5)
w k (2 + q ( n
2))
q  j6=k w j
l(2 q )(2 + q (n 1))
wk (2 + q (n 2)) q  j6=k w j
,
,
(2 q )(2 + q (n 1))
(wk (2 + q (n 2)) q  j6=k w j )2
Nash
PCournot,k
=
.
l(2 q )2 (2 + q (n 1))2
In the Nash equilibrium under Bertrand competition firm k maximizes Pk with respect
to its price pk given the other firms’ prices p
k.
Firm k’s response function can be cal-
culated as
pk ( p
k)
=
wk
2
q  j6=k ( w j
2(1 + q ( n
pj )
2))
.
Summing over all n firms yields
n
Â
k =1
pk =
Ânk=1 wk
2
q (n
1) Ânk=1 (wk pk )
,
2(1 + q (n 2))
which can be transformed using the previous Equation together with Equation (A.2) to
retrieve the Bertrand Nash equilibrium
(A.6)
Nash
q Bertrand,k
=
(1 + q ( n
2))(wk (q 2 (n2
5n + 5) + 3q (n
2) + 2)
q (1 + q ( n
2)) Â j6=k w j )
,
l(1 q )(1 + q (n 1))(2 + q (n 3))(2 + q (2n 3))
wk (q 2 (n2 5n + 5) + 3q (n 2) + 2) q (1 + q (n 2)) Â j6=k w j
Nash
p Bertrand,k =
,
(1 + q (n 1))(2 + q (n 3))(2 + q (2n 3))
(1 + q (n 2))(wk (q 2 (n2 5n + 5) + 3q (n 2) + 2) q (1 + q (n 2)) Â j6=k w j )2
Nash
P Bertrand,k
=
.
l(1 q )(1 + q (n 1))2 (2 + q (n 3))2 (2 + q (2n 3))2
230
A.1 Oligopoly competition
As Häckner (2000) shows, Nash prices are always higher under Cournot competition
than under Bertrand competition for substitute goods (q > 0). Instead, if goods are
complements (q < 0) and vertical differentiation between firms is high, Nash prices
of low-quality firms may be higher under Bertrand competition than under Cournot
competition. With respect to profits there are different nuances. For complementary
goods, Nash profits are always higher under Bertrand competition than under Cournot
competition. Instead, if goods are substitutes, the opposite holds unless vertical differentiation between firms is low, when Nash profits of high-quality firms may be higher
under Bertrand competition than under Cournot competition.
Collusive equilibrium In the collusive equilibrium firms employ JPM, i.e., firms behave like a single monopolist and maximize Ânk=1 Pk . Therefore, the collusive equilibrium is identical under Bertrand and Cournot competition. Using Equation (A.1) and
summing over the corresponding profit functions, joint profit of all n firms is given
by
n
n
k =1
k =1
 Pk =  ( wk qk )
Noting that
∂ Ânk=1 (qk  j6=k q j )
∂qk
n
l  q2k
k =1
n
lq  (qk  q j ).
k =1
j6=k
= 2 Â j6=k q j , the first-order condition of JPM can be calculated
as
qk (q
k)
=
wk
2lq  j6=k q j
2l
.
Again summing over all n firms results in
n
 qk =
k =1
Ânk=1 wk
2l
q (n
n
1) Â q k ,
k =1
231
Appendix A Theoretical Analyses
which finally yields the collusive equilibrium using the previous Equation and Equation (A.2) as
q JPM =
p JPM =
(A.7)
P JPM =
w k (1 + q ( n
2l(1
2))
q  j6=k w j
q )(1 + q (n
wk
,
2
w k ( w k (1 + q ( n
4l(1
1))
2))
q )(1 + q (n
,
q  j 6 =i w j )
1))
.
Note that JPM prices are linearly connected to vertical differentiation as firm k’s price
in collusive equilibrium depends solely on its own quality.
Symmetric firms In case of symmetric firms without vertical product differentiation,
i.e., wk = w, 8k 2 {1, ..., N }, k’s demand function, i.e., Equation (A.3), simplifies to
qk =
=
(w
pk )(1 + q (n
l (1
w
l (1 + q ( n
|
{z
G
=G
2))
q  j6=k ( w
pj )
q )(1 + q (n 1))
1 + q ( n 2)
p +
1)) l(1 q )(1 + q (n 1)) k l(1
} |
{z
}
|
L
 j6=k q j
Lpk + Q
n 1
q ( n 1)
q )(1 + q (n
{z
Q
 j6=k q j
1)) n 1
}
with G, L, Q > 0 for substitute goods (q > 0). Consequently, the Walrasian equilibrium
given by Equation (A.4), which predicts marginal cost pricing, simplifies to
qWalras =
w
l (1 + q ( n
pWalras = 0,
PWalras = 0.
232
1))
,
A.1 Oligopoly competition
In the Nash equilibrium under Cournot competition firm k maximizes Pk with respect
to qk . With symmetric firms, Equation (A.5) yields the Cournot Nash equilibrium
w
,
l(2 + q (n 1))
w
Nash
pCournot
=
,
2 + q ( n 1)
w2
Nash
PCournot
=
.
l(2 + q (n 1))2
Nash
qCournot
=
In the Nash equilibrium under Bertrand competition firm k maximizes Pk with respect
to pk . With symmetric firms and Equation (A.6), the Bertrand Nash equilibrium is given
by
w (1 + q (n 2))
,
l(2 + q (n 3))(1 + q (n 1))
w (1 q )
Nash
p Bertrand
=
,
2 + q ( n 3)
w 2 (1 q )(1 + q (n 2))
Nash
P Bertrand
=
.
l(2 + q (n 3))2 (1 + q (n 1))
Nash
q Bertrand
=
Finally, in the collusive equilibrium, with firms employing JPM and irrespective of
Bertrand or Cournot competition, Equation (A.7) simplifies to
w
2l(1 + q (n
w
p JPM = ,
2
w2
P JPM =
4l(1 + q (n
q JPM =
1))
1))
,
.
233
Appendix A Theoretical Analyses
Parameterized and scaled theoretical benchmarks
TABLE A.1: Scaled theoretical benchmarks of oligopoly competition for each treatment with
symmetric firm as displayed in the experiment.
Bertrand
Duopoly
p
Walras
q
Walras
P
Walras
=0
= 60.00
=0
q
= 100.00
P
Walras
=0
q Nash = 45.00
q Nash = 62.50
q
JPM
= 30.00
P JPM = 1875.00
P Nash = 1406.25
p JPM = 50.00
q JPM = 50.00
P JPM = 1500.00
pWalras = 0
pWalras = 0
qWalras = 42.86
qWalras = 100.00
PWalras = 0
PWalras = 0
p Nash = 16.67
p Nash = 30.00
q Nash = 35.71
q Nash = 70.00
P
Nash
= 1406.25
P Nash = 1406.25
p JPM = 50.00
p JPM = 50.00
q JPM = 21.43
q JPM = 50.00
P
JPM
= 2531.42
P JPM = 1674.22
pWalras = 0
pWalras = 0
qWalras = 33.33
qWalras = 100.00
P
Walras
=0
PWalras = 0
p Nash = 12.50
p Nash = 25.00
q Nash = 29.17
q Nash = 75.00
P Nash = 1406.25
JPM
P Nash = 1406.25
= 50.00
p JPM = 50.00
q JPM = 16.67
q JPM = 50.00
p
P JPM = 3214.29
234
=0
Walras
p Nash = 37.50
p JPM = 50.00
Quadropoly
p
p Nash = 25.00
P Nash = 1406.25
Triopoly
Cournot
Walras
P JPM = 1875.00
A.1 Oligopoly competition
TABLE A.2: Scaled theoretical benchmarks of oligopoly competition for the asymmetric
Bertrand treatments as displayed in the experiment.
Incumbent
p
Walras
q
Walras
P
Walras
p
Triopoly
Nash
=0
= 55.92
=0
= 18.26
q Nash = 41.52
P Nash = 2109.38
p
=0
q
Walras
= 37.63
P
Walras
=0
p
Nash
= 15.82
q Nash = 33.90
P Nash = 1406.25
p JPM = 50.00
p JPM = 50.00
q JPM = 27.96
q JPM = 18.82
P
JPM
= 4897.86
pWalras = 0
q
Quadropoly
Entrant
Walras
Walras
= 44.50
P JPM = 3166.26
pWalras = 0
qWalras = 30.14
PWalras = 0
PWalras = 0
p Nash = 14.00
p Nash = 11.98
q Nash = 34.23
q Nash = 27.95
P Nash = 2109.38
P Nash = 1406.25
= 50.00
p JPM = 50.00
q JPM = 22.25
q JPM = 15.07
p
JPM
P JPM = 3889.00
P JPM = 2466.92
TABLE A.3: Nash predictions p Nash as measured by Walrasian-based degree of tacit collusion
jWalras under asymmetric Bertrand competition controlling for scaling in each
treatment.
Triopoly
Quadropoly
Incumbent
Entrant
0.37
0.28
0.32
0.24
235
Appendix A Theoretical Analyses
A.2 Margin squeeze regulation
Retail demand for firm k 2 { A, B, D } in case of n = 3 active firms according to Shubik
and Levitan (1980) is given by
Triopoly
qk
=
1
· (1
3
pk
p A + pB + pD
).
3
g · ( pk
Assume w.l.o.g. that Firm A is the single wholesale access provider to the retailer
Firm D. Firms’ profits are then given by
p A = p A · q A ( p A , p B , p D ) + a · q D ( p A , p B , p D ),
p B = p B · q B ( p A , p B , p D ),
pD = ( pD
a ) · q D ( p A , p B , p D ).
Three-stage model with a competitive fringe
The subgame-perfect Nash equilibria
are determined through backward induction. Under NR, in Stage 3, Firm D’s first order
condition
∂p D
∂p D
NR =
= 0 yields its best response p D
1 2ag+gp A +gp B +3a+3
2
3+2g
firms’ prices. In Stage 2, integrated firms i 2 { A, B} solve
NR )
∂pi ( p D
∂pi
given integrated
= 0 simultaneously,
thus yielding
22ag4 + 101ag3 + 150ag2 + 95g3 + 72ag + 384g2 + 504g + 216
,
57g4 + 428g3 + 1092g2 + 1152g + 432
16ag4 + 64ag3 + 84ag2 + 95g3 + 36ag + 384g2 + 504g + 216
p BNR =
.
57g4 + 428g3 + 1092g2 + 1152g + 432
p NR
A =
In Stage 1, Firm A chooses the wholesale price a NR anticipating its competitors’ best
responses, i.e., according to
a NR =
236
NR NR
∂p A ( p NR
A ,p B ,p D )
∂a
= 0, which yields
4 49g4 + 336g3 + 828g2 + 864g + 324 19g2 + 54g + 36
.
973g7 + 17071g6 + 111816g5 + 370476g4 + 686880g3 + 723168g2 + 404352g + 93312
A.2 Margin squeeze regulation
Simple plugging in of equilibrium prices gives equilibrium retail demands and equilibrium profits dependent on g. For the sake of brevity, the algebraic expressions are not
denoted here, but can be found in the online appendix. Testing for the margin squeeze
condition
D = p NR
A
=
a NR
g2 665g4 + 3543g3 + 6984g2 + 6048g + 1944
973g7 + 17071g6 + 111816g5 + 370476g4 + 686880g3 + 723168g2 + 404352g + 93312
shows that D < 0 for g > 0 and thus, Firm A engages in a margin squeeze if retail goods
are substitutes. Despite that, Firm D is still active in the retail market with a positive
demand
NR
qD
=
1 399g7 + 9893g6 + 67488g5 + 218268g4 + 388368g3 + 391392g2 + 209952g + 46656
6 973g7 + 17071g6 + 111816g5 + 370476g4 + 686880g3 + 723168g2 + 404352g + 93312
as Firm A does not find it profitable to foreclose its downstream competitor, which is
shown in the following.
Foreclosure occurs if Firm A’s wholesale price is so high compared to retail prices such
that Firm D is unable to set a retail price that would yield a positive profit. Then, Firm D
does not supply any retail consumers and a retail duopoly ensues with
qiForeclosure =
1+g
(1
3
pi
g
(2
(3 + 2g)
( p A + p B ))
as integrated firm i’s retail demand (Höffler, 2008). Profit-maximization of integrated
firms in case of foreclosure yields equilibrium profits piForeclosure (g). Figure A.1 depicts
the difference of equilibrium profits piForeclosure
piNR for both integrated firms. Irre-
spective of the degree of differentiation, for substitute goods, i.e., g > 0, Firm B seeks
foreclosure due to higher downstream profits, but Firm A does not. Instead, the Firm A
benefits from a viable wholesale offer to Firm D and thus has no incentive to foreclose
its competitor even in the absence of regulation.
237
Appendix A Theoretical Analyses
F IGURE A.1: Comparison of profits in the case of (non-)foreclosure under NR.
Under MSR, in Stage 2, Firm A is constrained in its retail price setting by the requirement that D
0. Because it prefers to engage in a margin squeeze under NR, the margin
squeeze condition is binding and p MSR
= a MSR . Firm B’s corresponding retail price is
A
p BMSR =
1 7ag2 +9ag+15g+18
.
2
7g2 +24g+18
In Stage 1, Firm A sets a MSR =
19g2 +54g+36
3
2 7g3 +81g2 +180g+108 .
Again,
equilibrium retail demands and equilibrium profits can be obtained by simple computations.
Two-stage model with simultaneous retail pricing
The initial three-stage game is re-
duced to a two-stage game in which all three firms choose retail prices simultaneously.
Under NR, if Firm A makes a viable wholesale offer in Stage 1, in Stage 2 firms choose
retail prices
1
2
3
p BNR =
2
1
NR
pD
=
2
p NR
A =
238
5ag2 + 9ag + 15g + 18
,
5g2 + 21g + 18
ag2 + ag + 5g + 6
,
5g2 + 21g + 18
7ag2 + 21ag + 18a + 15g + 18
.
5g2 + 21g + 18
A.2 Margin squeeze regulation
Anticipating these retail prices, Firm A chooses the profit-maximizing wholesale
price
a NR =
3(25g3 + 120g2 + 198g + 108)
.
20g4 + 249g3 + 909g2 + 1296g + 648
If instead, Firm A does not make a viable wholesale offer but decides to foreclose
Firm D, integrated firms’ profits are given by piForeclosure (g) as obtained for the preForeclosure > p NR for the degree of substitutability
vious three-stage model. Solving p A
A
yields g > 26.77 =: g, i.e., Firm A benefits from foreclosure if retail goods are close substitutes (see Figure A.2). Therefore, as noted by Atiyas et al. (2015) and Bourreau et al.
(2011), foreclosure is the unique equilibrium outcome for g > g. Otherwise, Firm A
makes a viable wholesale offer and all three firms participate in the downstream market in equilibrium.
F IGURE A.2: Comparison of Firm A’s profit in the case of (non-)foreclosure under NR.
While in case of foreclosure Firm A’s price structure clearly violates the margin squeeze
constraint, an additional analysis is warranted for g  g. Testing for the margin squeeze
condition
D = p NR
A
a NR =
g 5g2 9g 18
3
2 20g4 + 249g3 + 909g2 + 1296g + 648
239
Appendix A Theoretical Analyses
demonstrates that D < 0 for g > 3 so that only for strongly differentiated goods, i.e.,
g 2 [0, 3], the margin squeeze condition is not binding. Therefore, under MSR, in Stage
2 Firm A sets p MSR
= a MSR for all g > 3. The competitors’ prices for g 2 (3, g) are given
A
by
1 7ag2 + 9ag + 15g + 18
,
3
5g2 + 16g + 12
1 13ag2 + 30ag + 18a + 15g + 18
MSR
pD
=
.
3
5g2 + 16g + 12
p BMSR =
In Stage 1, Firm A anticipates these decisions and sets the wholesale price a MSR =
3 5g+6
4 g2 +9g+9 .
Two-stage model with simultaneous retail quantity competition In order to examine the effect of MSR in case of retail quantity competition, a two-stage game that is
similar to the one detailed above is considered with the exception that firms choose
quantities in Stage 2 and compete according to the demand structure suggested by
Singh and Vives (1984). Following the generalization by Häckner (2000) for more than
two firms, inverse retail demand of firm k 2 { A, B, D } in the case of symmetric competitors is given by
pk = w
l qk + q  q j
j6=k
!
with w, l > 0 and q as a standardized measure of substitutability (see Appendix A.1).
Here only the case of q 2 [0, 1] is considered. For ease of illustration, let w = 100 and
l = 1.
Under NR, in Stage 2, firms simultaneously choose
1 aq 100q + 200
,
2
q2 q 2
1 aq + 2a + 100q 200
NR
,
qD
=
2
q2 q 2
qiNR =
240
A.2 Margin squeeze regulation
where i 2 { A, B}. In Stage 1, Firm A anticipates these decisions and sets
a NR =
100(q 3 4q 2 + 2q + 4)
.
2q 3 + 3q 2 8q 8
In order to identify whether Firm A engages in a margin squeeze, Firm A’s retail price
p NR
A that arises from all firms’ quantity decisions is calculated and then
D = p NR
A
a NR =
50q 2q 2 5q + 2
2q 3 + 3q 2 8q 8
is determined to test the margin squeeze condition, for which D < 0 if q 2 (0, 0.5). Therefore, Firm A engages in a margin squeeze for rather differentiated goods.
Plugging in equilibrium quantities and the optimal wholesale price yields equilibrium profits pkNR , which are now compared to the case of foreclosure which ensues
profits piForeclosure (q ). Provided that Firm D is foreclosed, the integrated firms choose
qiForeclosure =
100
q +2 .
Comparing profits between foreclosure and non-foreclosure for both
integrated firms yields
Foreclosure
pA
NR
pA
=
p BForeclosure
p BNR =
Foreclosure
where p A
2500(q 4 + 4q 3
4q 2
16q + 16)
,
(q + 2) (2q 3 + 3q 2 8q 8)
2500q 7q 5 + 24q 4 36q 3 128q 2 + 48q + 128
2
(q + 2)2 (2q 3 + 3q 2
NR < 0 and p Foreclosure
pA
B
8q
8)
2
,
p BNR > 0 for all q 2 [0, 1]. Hence, in analogy
to the two-stage price competition model, Firm B prefers foreclosure, whereas Firm A
benefits from making Firm D a viable wholesale offer.
241
Appendix A Theoretical Analyses
Under MSR, in Stage 2, constrained pricing of Firm A and simultaneous profit maximization by its competitors yields retail quantities
2( a + 50q 100)
,
2q 2 q 2
2aq 2 3aq 100q 2 + 300q 200
q BMSR =
,
(q 2) (2q 2 q 2)
2( aq + 50q 2 a 150q + 100)
MSR
qD
=
.
(q 2) (2q 2 q 2)
q MSR
=
A
In Stage 1, Firm A sets a MSR =
25q + 50. Ensuing prices and profits allow to com-
pute consumer surplus, producer surplus, and total surplus as in the main analysis.
Figure A.3 depicts the effect of MSR relative to NR for these welfare measures. The
implications are similar to the investigated three-stage price competition model: while
MSR is to consumers’ detriment, producers benefit from it. Because the former effect
outweighs the latter, total surplus is unambiguously lower under MSR than NR.
F IGURE A.3: MSR effect on consumer surplus, producer surplus, and total surplus.
242
A.3 Wholesale competition
A.3 Wholesale competition
The market structure and retail demand is identical to the previous analysis of infrastructure competition under margin squeeze regulation in Appendix A.2. Thus, retail
demand in the case of n = 3 active firms is given by
Triopoly
qk
=
1
· (1
3
pk
g · ( pk
p A + pB + pD
),
3
and in the case for foreclosure (n = 2) by:
qkForeclosure =
1+g
· (1
3
pk
g
· (2
(3 + 2g)
( p A + p B )).
In the following, let g = 30. Assume w.l.o.g. that Firm A provides wholesale access to
the retailer Firm D. This yields profits
p A = p A · q A ( p A , p B , p D ) + a · q D ( p A , p B , p D ),
p B = p B · q B ( p A , p B , p D ),
pD = ( pD
a ) · q D ( p A , p B , p D ).
In the following, Nash predictions are calculated for the market scenarios: (i) wholesale
monopoly under no regulation, (ii) wholesale monopoly under margin squeeze regulation, (iii) wholesale competition under no regulation, and (iv) wholesale competition
under margin squeeze regulation, each for all four timing models as described in Subsection 5.2.1. For the sequential-move Timing Models (1), (2), and (4) subgame-perfect
Nash equilibria are determined through backward induction. In order to facilitate the
comparison of theoretical predictions and experimental results, final prices and profits
are scaled as in the experiment. Note that scaling affects only the output, but calcula-
243
Appendix A Theoretical Analyses
tions are based on the original Shubik and Levitan (1980) values. Price (profit) values
are multiplied by factor q = 100/0.15 (Q = 400) to obtain scaled values.
Timing Model (1): Under a wholesale monopoly, in stage II, two (integrated) or three
firms may operate in the retail market depending on the wholesale price chosen in stage
I. In stage II, in case of foreclosure, i.e., q D = 0, integrated firms’ profit-maximizing
prices are given by p Foreclosure
= p BForeclosure = 55.55 and profits amount to p AForeclosure =
A
p BForeclosure = 15.04. Instead, in case there is a viable wholesale offer, firms choose reTriopoly
tail prices p A
1
= q ( 22
+
265
572 a ),
Triopoly
pB
1
= q ( 22
+
155
572 a ),
Triopoly
and p D
1
= q ( 22
+
193
286 a ).
In stage I, anticipating retail prices, Firm A chooses the monopolistic wholesale price
Triopoly
= 61.05, p B
Triopoly
= 14.69, and p D
a Triopoly = 66.36. Ensuing retail prices are given by p A
Triopoly
pD
Triopoly
= 75.08 and profits by p A
Triopoly
Foreclosure to p
Comparing p A
A
= 14.97, p B
Triopoly
= 48.29, and
Triopoly
= 0.48.
, Firm A prefers the foreclosure outcome and thus sets
a wholesale price a Foreclosure 2 (93.19, 100], which forces the retailer to exit the retail market.
Taking into account margin squeeze regulation, foreclosure is ruled out as a valid marTriopoly
ket outcome and therefore does not constitute an equilibrium. However, as a A
Triopoly
pA
>
, Firm A would violate the margin squeeze condition if it could set its prices
freely. Instead, Firm A is required to maximize its profit p A subject to the condition
a A  p A in stage II, while the other firms maximize profits unconstrained. This yields
1
p MSR
= a, p BMSR = q ( 32
+
A
365
832 a ),
MSR = q ( 1 +
and p D
32
701
832 a ).
In stage I, Firm A sets the
monopoly wholesale price to a MSR
= 66.16 and corresponding retail prices are given by
A
MSR = 76.57 with profits p MSR = 15.09, p MSR = 15.66,
p MSR
= 66.16, p BMSR = 49.86, and p D
B
A
A
MSR = 0.68.
and p D
Considering unregulated wholesale competition, two equilibria emerge, namely a competitive and a foreclosure type. Atiyas et al. (2015) show that for g > 26.77 (and observable wholesale contracts) the foreclosure outcome constitutes an additional Nash
244
A.3 Wholesale competition
TABLE A.4: Theoretical predictions for Timing Model (1).
No regulation
Margin squeeze regulation
Wholesale monopoly
aA =
pA =
pB =
pD =
100.00
55.56
55.56
100.00
aA =
pA =
pB =
pD =
66.19
66.19
49.86
76.57
Wholesale competition
a A = aB =
p A = pB =
pD =
100.00
55.56
100.00
a A = aB =
0.00
p A = p B = p D = 30.30
or:
a A = aB =
0.00
p A = p B = p D = 30.30
equilibrium next to the competitive outcome. As shown for the case of a wholesale
monopoly, an integrated firm does not find it profitable to deviate from the state of coordinated foreclosure in the wholesale market, because wholesale profits are outweighed
by the retailer’s business stealing effect in the retail market, even at the monopoly
price. Moreover, given the nonviable wholesale offers, no firm k has an incentive to
deviate from its foreclosure price pkForeclosure . In contrast, as shown by Bourreau et al.
(2012b), if a firm is required to make a viable wholesale offer, integrated firms always find it profitable to undercut their rival in the wholesale market for g < 40.97.
Competitive
Once wholesale prices are driven to zero, i.e., a A
Competitive
= aB
= 0, firms can-
not unilaterally increase the wholesale price profitably. Thus the competitive outcome
Competitive
aA
Competitive
pk
Competitive
= aB
Competitive
= 0 with ensuing retail prices pk
= 30.30 and profits
= 5.79, 8k 2 { A, B, D }, constitutes a Nash equilibrium.
Whereas in the case of no regulation two types of equilibria coexist, the competitive
equilibrium is unique in the case of wholesale competition under margin squeeze regulation. Integrated firms are now unable to foreclose the retailer, due to the margin
squeeze condition, while the Bertrand logic applies as described above for the unregulated case. These results are summarized in Table A.4.
245
Appendix A Theoretical Analyses
Timing Model (2): In stage III, the retailer’s optimal price as a follower is given by
1
its best response function, i.e., p D = BR( p A , p B ) = q ( 42
+
5
21 p A
+
5
21 p B
+ 12 a) across
all scenarios. Anticipating the retailer’s reaction, integrated firms’ simultaneous (un13
constrained) profit-maximization in stage II yields p A = q ( 261
+
13
q ( 261
+
5320
15109 a )
and p B =
7595
30218 a ).
In the monopoly case, Firm A maximizes its profit by setting its wholesale price to
Triopoly
a A = 67.39 in stage I. Ensuing retail prices are given by p A
Triopoly
50.14, and p D
Triopoly
and p D
Triopoly
= 75.07 and firms make profits p A
Triopoly
= 56.94, p B
Triopoly
= 15.61, p B
Triopoly
= 0.37. Note that under this timing model p A
=
= 14.05,
exceeds the foreclo-
Foreclosure , because Firm A internalizes the retailer’s reaction to its own
sure profit p A
prices. Therefore, Firm A finds it always profitable to make a viable wholesale offer
to Firm D.
Although foreclosure does not constitute an equilibrium under no regulation, Nash
prices of Firm A still violate the margin squeeze condition.
Taking into account
this condition, constrained maximization of Firm A’s profit in stage II yields prices
13
p MSR
= a A and p BMSR = q ( 391
+ 365
A
782 a ). The optimal wholesale price in stage I is given by
MSR = 80.72.
a MSR = 70.14 and respective retail prices are p MSR
= 70.14, p BMSR = 54.90, p D
A
MSR = 16.24, p MSR = 16.84, and p MSR = 0.70. Note that
Firms’ profits amount to p A
B
D
margin squeeze regulation leads to unambiguously higher prices and increased profits
compared to the no regulation outcome, given this timing model.
In the case of wholesale competition, the Bertrand logic, as laid out in Timing Model (1),
applies equally with regard to the integrated firms’ behavior in stage I. Moreover, the
competitive outcome is unique, because one of the integrated firms will always find
it profitable to unilaterally deviate from coordinated foreclosure and supply the retailer.
246
A.3 Wholesale competition
TABLE A.5: Theoretical predictions for Timing Model (2).
No regulation
Wholesale monopoly
aA =
pA =
pB =
pD =
67.39
56.94
50.15
75.07
Wholesale competition
a A = aB =
0.00
p A = p B = p D = 30.30
Margin squeeze regulation
70.14
70.14
54.90
80.72
aA =
pA =
pB =
pD =
a A = aB =
0.00
p A = p B = p D = 30.30
Given the theoretical prediction, margin squeeze regulation does not affect the market
outcome under wholesale competition in Timing Model (2), because equilibrium prices
in the unregulated outcome do not violate the margin squeeze constraint. See Table A.5
for a summary of results.
Timing Model (3): In the case of an unregulated monopolistic wholesale provider, simultaneous setting of all prices (wholesale and retail) leads to foreclosure as the unique
equilibrium. Consider, in contrast, a situation in which Firm D makes positive profit,
i.e., a A < p D . Obviously, Firm A can then increase its profit by setting a A = p D . However, Firm D would in turn increase its retail price p D as long as it is able to obtain a
positive demand (q D > 0). Consequently, this reverse Bertrand logic gives rise to foreclosure as the unique equilibrium.
In the case of margin squeeze regulation, Firm A solves
∂p A
∂p A
= 0 and
∂p A
∂a A
= 0 simulta-
neously subject to the constraint a A  p A , while Firm B and Firm D solve first order
conditions
∂p B
∂p B
= 0 and
∂p D
∂p D
= 0. Optimal prices are given by a MSR
= p MSR
= 67.61,
A
A
MSR = 77.80 leading to profits p MSR = 15.08, p MSR = 16.06, and
p BMSR = 50.49, and p D
B
A
MSR = 0.65.
pD
If both integrated firms are active in the wholesale market, the same rationale as under Timing Model (1) applies, i.e., the competitive as well as the foreclosure outcome
constitute an equilibrium. On the one hand, if both firms choose a wholesale price that
247
Appendix A Theoretical Analyses
TABLE A.6: Theoretical predictions for Timing Model (3).
No regulation
Margin squeeze regulation
Wholesale monopoly
aA =
pA =
pB =
pD =
100.00
55.56
55.56
100.00
aA =
pA =
pB =
pD =
67.61
67.61
50.49
77.80
Wholesale competition
a A = aB =
p A = pB =
pD =
100.00
55.56
100.00
a A = aB =
0.00
p A = p B = p D = 30.30
or:
a A = aB =
0.00
p A = p B = p D = 30.30
forecloses Firm D, there is no unilateral deviation that increases an integrated firm’s
profit, because the business stealing effect outweighs the wholesale revenue effect. On
the other hand, in the case of the competitive outcome, an integrated firm is unable to
establish the foreclosure outcome unilaterally.
If wholesale competition is combined with margin squeeze regulation, the foreclosure
equilibrium disappears—as under Timing Model (1)—and the competitive outcome
constitutes the unique equilibrium. These results are summarized in Table A.6.
Timing Model (4): Like in Timing Model (2), the retailer as a follower maximizes
its profit given the previously set wholesale price(s) and retail prices of the integrated
1
firms, i.e., according to its best response function it chooses p D = BR( p A , p B ) = q ( 42
+
5
21 p A
+
5
21 p B
+ 12 a). In stage I, integrated firms maximize profits simultaneously taking
Triopoly
into account the reaction by Firm D in stage II. Optimal prices are given by a A
Triopoly
pA
Triopoly
= 51.25, p B
Triopoly
obtain profits p A
Triopoly
= 46.09, and consequently p D
Triopoly
= 15.07, p B
= 64.67. Accordingly, firms
Triopoly
= 11.86, and p D
=
= 1.14. Again, considering
the retailer as a follower allows the integrated firms to internalize Firm D’s reaction to
their own prices which makes foreclosure relatively less profitable.
248
A.4 Coopetition in online advertising markets
TABLE A.7: Theoretical predictions for Timing Model (4).
No regulation
Wholesale monopoly
aA =
pA =
pB =
pD =
Wholesale competition
a A = aB =
0.00
p A = p B = p D = 30.30
Margin squeeze regulation
51.25
51.25
46.09
64.68
51.25
51.25
46.09
64.68
aA =
pA =
pB =
pD =
a A = aB =
0.00
p A = p B = p D = 30.30
The margin squeeze condition is non-binding, because Firm A’s equilibrium prices in
the case of an unregulated wholesale monopoly do not constitute a margin squeeze.
Therefore, the theoretical prediction is the same as under an unregulated wholesale
monopoly.
Having ruled out foreclosure as an equilibrium in the case of a wholesale monopoly, the
same rationale holds under unregulated wholesale competition, because an integrated
firm has always an incentive to deviate in the upstream market and charge the monopolistic wholesale price. Thus, as argued above, the competitive outcome remains as the
unique equilibrium.
In consequence, under wholesale competition the margin squeeze regulation does
not affect the theoretical prediction and the outcome is identical to the unregulated
Competitive
wholesale competition scenario, i.e., a A
Competitive
pk
Competitive
= aB
Competitive
= 0, pk
= 30.30, and
= 5.79, 8k 2 { A, B, D }. See Table A.7 for a summary of results.
A.4 Coopetition in online advertising markets
Prices and profits of special-interest CP given social login adoption decisions
Scenario (b,b) – None of the CPs adopts the social login: In this case market shares are
given by D bA = DBb = 1/2 and equilibrium profits, which equal equilibrium prices,
249
Appendix A Theoretical Analyses
are obtained by simultaneously solving (7.3) and (7.4). Note that neG(s) ( Dsb , d, aG(s) ) ⌘
nG(s) ( Dsb , d, 0, aG(s) ) and nes ( Dsb , d, as ) = ns ( Dsb , d, as , 0), i.e. view-throughs under exclu-
sive advertising can be calculated by assuming a rival CP with targeting rate a = 0.
Thus, for the base scenario (b,b), conditions (7.3) and (7.4) can be stated as
nG(s) ( Dsb , d, as , aG(s) )
b
Pb,b
G + ns ( Ds , d, as , a G (s) )
e
b
Pb,b
s = n G (s) ( Ds , d, a G (s) )
nG(s) ( Dsb , d, as , aG(s) )
b
Pb,b
G + ns ( Ds , d, as , a G (s) )
e
b
Pb,b
s = ns ( Ds , d, as )
Pb,b
G
Pb,b
s .
Solving simultaneously yields special-interest CP s’s equilibrium price and profit of
b,b
pb,b
s = Ps =
1 b
d a · (2 (1
2 s
abG ) + d (2 abG
as )).
Scenario (l,b) – CP s adopts the social login, but its rival does not: If CP s adopts the social
login, but its special-interest rival
s does not, it gains a relative advantage with regard
to the base utility that users derive from consuming its services, i.e. uls = ub + q with
q
0. In consequence, it is able to increase its market share, as demand is now given by
Dsl =
t +q
2t ,
whereas its rival experiences demand D l s =
rate is increased to als = min{fs abs , 1} with fs
t q
2t .
Moreover, CP s’s targeting
1. On the other hand, adoption of
the social login also increases the targeting rate of CP G for the mass of consumers
that consume services of CP s, i.e. alG(s) = min{fG abG(s) , 1} with fG
1. Note that the
targeting rate of CP G with respect to users that consume services of the rival CP
unaffected, i.e. alG(
s)
s is
= abG . Solving (7.3) and (7.4) analogous to scenario (b,b) yields
price and profit of
l,b
pb,b
s = Ps =
d als (t + q )
· 2
2t
d als
2 alG (1
d) .
Scenario (b,l) – CP s does not adopt the social loing, but its rival does: This case is symmetric
to case (l,b), but CP s is now put at a disadvantage with regard to competition between
250
A.4 Coopetition in online advertising markets
the special-interest outlets as its rival benefits from consumers’ increased valuation for
its services due to the adoption of the social login. Whereas this diminishes its market share to Dsl =
t q
2t ,
its own advertising rate together with the targeting rate of the
general-interest CP for its mass of consumers Ds is the same as in the base scenario, i.e.
als = abs and alG(s) = abG(s) . Solving (7.3) and (7.4) analogous to scenario (b,b) yields price
and profit of
b,l
pb,l
s = Ps =
d abs (t q )
· 2 (1
2t
abG )
d (abs
2 abG ) .
l,b
Note that Pb,l
s is symmetric to Ps with regard to the market share
t +q
2t ,
given that
targeting rates differ by the factor fk .
Scenario (l,l) – Both CPs adopt the social login: If both CPs adopt the social login, users of
both outlets benefit from an increased base utility when consuming the services of their
respective CP, i.e., uls = ub + q. However, due to symmetry, CPs do not gain any competitive advantage and the surplus is appropriated exclusively by consumers. Thus,
market shares are symmetric and identical to the base scenario (b,b), Dsl = Dsb = 12 .
On the contrary, the impact of the social login on targeting rates is still effective, i.e.
als = fs abs for both s = A, B and alG( A) = alG( B) = fG abG(s) . Solving (7.3) and (7.4) analogous to scenario (b,b) yields price and profit of
l,l
pl,l
s = Ps =
1 l
d a · (2 (1
2 s
alG ) + d (2 alG
als )).
Prices and profits of the general-interest CP given social login adoption
decisions
Scenario (b,b) – None of the CPs adopts the social login: If none of the CPs adopts the
social login, the targeting rates that determine the general-interest CP’s advertising
profit are given by the initial and symmetric targeting rates of special-interest CPs A
251
Appendix A Theoretical Analyses
and B, i.e., abA = abB = abs , and by the general-interest CP’s own initial targeting rate
abG( A) = abG( B) = abG for users of both CP A and B. Solving (7.3) and (7.4) as shown for
the profit-maximization of special-interest CPs yields CP G’s equilibrium price of
pb,b
= abG (1
G (s)
d ) · (1
1 b
a
2 G
d (abs
1 b
a ),
2 G
b,b
resulting in an equilibrium profit of Pb,b
G = 2p G (s) .
Scenarios (l,b) and (b,l) – One CP adopts the login, the rival does not: If one special-interest
CP adopts the social login, assume CP A, whereas the other special-interest CP does
not, assume CP B, targeting rates are asymmetrically affected by the social login. The
mass of users D A , i.e., users that choose to visit CP A, can now be targeted at both outlets CP A and CP G with targeting rates a A = alA = fs abs and aG( A) = alG( A) = fG abG(s) ,
respectively. In contrast, the targeting ability does not change with regard to DB , i.e.,
users that choose to visit CP B. Thus CP B and CP G can target these users according to their initial targeting rates a B = abs and aG( B) = abG(s) , respectively. Solving (7.3)
and (7.4) as shown for the profit-maximization of special-interest CPs yields CP G’s
equilibrium
(l,b)
(b,l )
(l,b)
(b,l )
p G ( A) = p G ( B) =
p G ( B) = p G ( A) =
1)(abG fG (fG (d 1)abG 2dfs abs + 2)
2t
b
b
q )((aG 2as )d abG + 2)(d 1)
,
2t
(t + q )(d
abG (t
and a total profit of
1 b
b,l
Pl,b
G = PG = t aG (d
q (fs fG
252
1) (
1
(d
2
2
2
1 ) ( fG
t + fG
q+t
1)) abs d + t( fG
1) t
q ( fG
q ) abG + ((fs fG + 1) t +
1)).
A.4 Coopetition in online advertising markets
Scenario (l,l) – Both CPs adopt the social login: If both special-interest CPs adopt the social
login, targeting rates are again symmetric with regard to CP A and CP B, but are now
increased by fs for special-interest CPs and fG for the general-interest CP, respectively.
Consequently, targeting rates in this case are given by a A = a B = als = fs abs and aG =
alG(s) = fG abG(s) . Solving (7.3) and (7.4) as shown for the profit-maximization of specialinterest CPs yields CP G’s equilibrium price of
pl,l
= (1
G (s)
d)alG · 1
d als
1
(1
2
d)alG ,
l,l
resulting in an equilibrium profit of Pl,l
G = 2p G (s) .
Comparative Statics
Critical threshold for adoption of the social login (CP s): Given CP s’s rationale to
adopt the login Dl,b
b,b
= Pl,b
s
Pb,b
s > 0, comparative statics with respect to the exoge-
nous model parameters demonstrate the following effects.
CP G’s targeting rate abG(s) :
∂Dl,b b,b
1
=
· ((fG fs
b
2t
∂aG(s)
1)t + q fG fs )(d
1)d abs < 0.
CP G’s increase in targeting rate fG :
∂Dl,b b,b
1
= · abG(s) abs (t + q )d (d
∂fG
t
1)fs < 0.
253
Appendix A Theoretical Analyses
CP s’s targeting rate abs :
∂Dl,b b,b
1
= · d (( abs fs 2 d + (d fG abG(s)
t
∂abs
+ ( abG(s) + abs )d + abG(s)
fG abG(s) + 1)fs
1)t + q fs (d fG abG(s)
fs abs d
fG abG(s) + 1)) < 0.
∂f
To show that the above condition is satisfied, solve ∂abs = 0 and reorder with respect to
p s 2
t ( fG 1 ) ( t + q ) ( t + q ) fG + t
abG(s) . As it is assumed that abG(s) > 0, abG(s),0 :=
is the unique
(d 1)((t +q )f2 t )
G
zeroing. It is easy to show that the effect of d on
abG(s),0
is positive. Therefore, d = 0 is
used to calculate a lower bound for abG(s),0 . As alG(s) < 1, solving abG(s),0
respect to t yields the unique solution t = 0. Any t > 0 yields abG(s),0
1/fG
1/fG
= 0 with
> 0, which
contradicts the assumption alG(s) = abG(s) · fG < 1.
CP s’s increase in targeting rate fs :
∂Dl,b b,b
1
= · d abs (t + q )(fG d abG(s)
∂fs
t
d abs fs
fG abG(s) + 1) > 0.
CP s’s increase in base utility q:
∂Dl,b
∂q
b,b
=
1
· d fs abs (2 fG d abG(s)
2t
d abs fs
2 fG abG(s) + 2) > 0.
Screen attention d:
∂Dl,b
∂d
b,b
=
2
· (( 1/2 abs fs 2 d + (1/2 + fG (d
t
+ 1/2 abs d
1/2)abG(s) )fs + ( d + 1/2)abG(s)
1/2)t + q ( 1/2 fs abs d + 1/2 + fG (d
To show that the above condition is satisfied, solve
∂fs
∂d
1/2)abG(s) )fs )abs > 0.
= 0 and reorder with respect to
t. The sign of the derivative changes twice: first at t = 0, second at t < 0 as long as
abs < 2 f
254
fG abG(s) 1
b
b
G a G (s) + a G (s)
2
, which is always satisfied under the assumption fG > 1.
A.4 Coopetition in online advertising markets
Critical threshold for social login offer (CP G): Given CP G’s rationale to offer the
login Ul,l
b,b
= Pl,l
G
Pb,b
G , comparative statics yield that parameters q and t do not
affect CP G’s profit. The effects of the remaining parameters are as follows.
CP G’s targeting rate abG(s) :
∂Ul,l b,b
=
∂abG(s)
⇣
2 (abG(s) fG 2
fs fG abs + abs
abG(s) fG 2 + fG + abG(s)
abG(s) )d
⌘
1 (d
1)
< 0.
To show that the above condition is satisfied, determine
2
ously negative under the assumption fG > 1 , fG
∂fes
.
∂abG(s)
The effect is unambigu-
1 > 0.
CP G’s increase in targeting rate fG :
∂Ul,l b,b
=
∂fG
2 ((d
1)fG abG(s)
fs abs d + 1) (d
1)abG(s) > 0.
CP s’s targeting rate abs :
∂Ul,l b,b
= 2 abG(s) (d
∂abs
1) (fs fG
To show that the above condition is satisfied, solve
∂fes
∂abs
1)d < 0.
= 0 and reorder with respect to
abG(s) . Given the assumptions alG(s) < 1 and fG > 1, it is easy to show that the unique
zeroing is outside of the relevant parameter range.
CP s’s increase in targeting rate fs :
∂Ul,l b,b
= 2 abG(s) (d
∂fs
1) abs d fG < 0.
255
Appendix A Theoretical Analyses
Screen attention d:
∂Ul,l b,b
=
∂d
2 ( fG 2 ( d
1)
d + 1 abG(s)
+ (1 + ( 2 fs d + fs ) abs )fG
1 + (2 d
1) abs )abG(s) < 0.
The above condition is satisfied under the assumption fG > 1.
Critical threshold for profitability of the social login (CP s): Given CP s’s condition
for profitability Dl,l
b,b
= Pl,l
s
Pb,b
s , comparative statics yields similar effects to the
comparative statics on CP s’s rationale to adopt the login (with the exceptions of q and
t) as the threshold for profitability is a special case of the latter. Parameters q and t
do not have an impact on profitability. The effects of the remaining parameters are as
follows.
CP G’s targeting rate abG(s) :
∂Dl,l b,b
= abs d (fG fs
∂abG(s)
1)(d
1) < 0.
CP G’s increase in targeting rate fG :
∂Dl,l b,b
= abG(s) (d
∂fG
1)abs fs d < 0.
CP s’s targeting rate abs :
∂Dl,l b,b
= d ( abs fs 2 + fG abG(s) fs + abs
∂abs
256
abG(s) d
fG abG(s) fs + fs + abG(s)
1) < 0.
A.4 Coopetition in online advertising markets
CP s’s increase in targeting rate fs :
∂Dl,l b,b
= d abs (fG d abG(s)
∂fs
fs abs d
fG abG(s) + 1) > 0.
Screen attention d:
∂Dl,l b,b
= 2 ((fG fs
∂d
1)(d
1 b
)a
2 G (s)
1
(fs
2
1)(abs d fs + abs d
1))abs > 0.
257
Appendix B
Statistical Analyses
B.1 Oligopoly competition in continuous time
TABLE B.1: Average degrees of tacit collusion over the entire time horizon across treatments.
Treatment
Obs.
j̄ Nash
p/q
Nash
j̄P
j̄Walras
p/q
j̄Walras
P
DB2
12
0.832
(0.249)
0.806
(0.302)
0.916
(0.124)
0.951
(0.075)
DB3
12
0.605
(0.324)
0.611
(0.301)
0.737
(0.216)
0.827
(0.134)
DC2
12
0.627
(0.550)
0.437
(1.030)
0.907
(0.138)
0.965
(0.064)
DC3
12
0.397
(0.484)
0.249
(0.702)
0.759
(0.193)
0.880
(0.112)
RB2
12
0.769
(0.343)
0.712
(0.453)
0.884
(0.172)
0.928
(0.113)
RB3
11
0.539
(0.306)
0.491
(0.324)
0.693
(0.204)
0.774
(0.144)
RC2
12
0.789
(0.259)
0.688
(0.344)
0.947
(0.065)
0.980
(0.021)
RC3
13
0.386
0.186
(0.473)
(0.745)
Standard deviations in parentheses.
0.754
(0.189)
0.870
(0.119)
259
Appendix B Statistical Analyses
B.2 Wholesale competition and Open Access regulation
Integrated firms’ profit under unregulated wholesale monopoly
In order to investigate the effect of the access provider role in the case of a wholesale
monopoly the following quantile regression is ran based on observations at the firm
level:
pijt = b 0 + b Period · t + b Firm A · Firm A + eijt ,
where pijt denotes the profit of integrated firm i in market cohort j and period t. The
dummy variable Firm A denotes whether the integrated firm represents the monopolistic wholesale provider.
TABLE B.2: Quantile regression of integrated firms’ profit pi in the case of a wholesale
monopoly.
Baseline: Firm B in Wholesale Monopoly under No Regulation (WM-NR).
pi
Firm A
5.416⇤⇤⇤
(1.805)
Period
0.001
(0.001)
Constant
18.604⇤⇤⇤
(3.972)
Observations
57,600
Clustered standard errors (by market cohort) in parentheses.
⇤ p < 0.10, ⇤⇤ p < 0.05, ⇤⇤⇤ p < 0.01
260
B.2 Wholesale competition and Open Access regulation
Margin squeeze regulation in the case of a wholesale monopoly
TABLE B.3: Quantile regression of wholesale market prices am , retail market prices ym , intef
grated firms’ average profits p AB , retailer’s profits p D and consumer surplus CS
in the case of a wholesale monopoly.
Baseline: Wholesale Monopoly & No Regulation (WM-NR).
Covariate
Margin squeeze
regulation
Period
Constant
Observations
(1)
am
(2)
ym
(3)
p AB
(4)
pD
8.286
(15.569)
9.928
(8.354)
2.661
(1.786)
0.165
(0.784)
0.083
(0.083)
0.006
(0.007)
0.006⇤⇤⇤
(0.002)
0.001⇤⇤⇤
(0.001)
> 0.001
(< 0.001)
> 0.001⇤
(< 0.001)
60.290⇤⇤⇤
(11.192)
59.911⇤⇤⇤
(7.423)
15.112⇤⇤⇤
(1.991)
1.287⇤
(0.703)
0.338⇤⇤⇤
(0.096)
55,200
55,200
55,200
55,200
55,200
(5)
f
CS
Clustered standard errors (by market cohort) in parentheses.
⇤ p < 0.10, ⇤⇤ p < 0.05, ⇤⇤⇤ p < 0.01
Pairwise treatment effects
TABLE B.4: Pairwise treatment effects on the wholesale market price am .
No Regulation
Wholesale Monopoly
Wholesale Competition
WC /w Price Commitment
Wholesale Monopoly
⇤
73.573
^⇤
86.085
_⇤⇤⇤
40.540
^⇤⇤⇤
73.573
Margin Squeeze Regulation
⇠
>⇤
⇠
72.572
⇠
83.082
_⇤⇤
49.049
^⇤⇤
72.572
p < 0.10, ⇤⇤ p < 0.05, ⇤⇤⇤ p < 0.01
261
Appendix B Statistical Analyses
TABLE B.5: Pairwise treatment effects on the retail market price ym .
No Regulation
Wholesale Monopoly
65.499
^⇤⇤
Wholesale Competition
88.407
_⇤⇤⇤
WC /w Price Commitment
49.560
Wholesale Monopoly
^⇤⇤
65.499
⇤
Margin Squeeze Regulation
⇠
⇠
<⇤
83.124
^⇤⇤
92.281
_⇤⇤⇤
67.415
⇠
83.124
p < 0.10, ⇤⇤ p < 0.05, ⇤⇤⇤ p < 0.01
TABLE B.6: Pairwise treatment effects on the average profit of integrated firms p AB .
No Regulation
Wholesale Monopoly
16.383
^⇤⇤
22.434
_⇤⇤⇤
12.243
^⇤⇤⇤
16.383
Wholesale Competition
WC /w Price Commitment
Wholesale Monopoly
⇤
Margin Squeeze Regulation
⇠
⇠
<⇤
20.750
⇠
22.802
_⇤⇤⇤
15.866
⇠
20.750
p < 0.10, ⇤⇤ p < 0.05, ⇤⇤⇤ p < 0.01
TABLE B.7: Pairwise treatment effects on the retailer’s profit p D .
No Regulation
Wholesale Monopoly
Wholesale Competition
WC /w Price Commitment
Wholesale Monopoly
⇤
262
p < 0.10, ⇤⇤ p < 0.05, ⇤⇤⇤ p < 0.01
0.899
⇠
0.258
^⇤⇤⇤
2.491
⇠
0.899
Margin Squeeze Regulation
⇠
<⇤⇤⇤
⇠
1.028
^⇤⇤
2.339
⇠
2.322
_⇤⇤⇤
1.028
B.2 Wholesale competition and Open Access regulation
f
TABLE B.8: Pairwise treatment effects on consumer surplus CS.
No Regulation
Wholesale Monopoly
0.292
WC /w Price Commitment
Wholesale Monopoly
⇤
0.153
⇠
_⇤
0.097
^⇤⇤⇤
0.461
_⇤⇤⇤
0.292
Wholesale Competition
Margin Squeeze Regulation
_⇤⇤
0.062
^⇤⇤⇤
0.298
_⇤
0.153
⇠
>⇤⇤
p < 0.10, ⇤⇤ p < 0.05, ⇤⇤⇤ p < 0.01
Analysis based on mean values
TABLE B.9: Average market prices, profits and consumer surplus per treatment.
Treatment
Markets
WM-NR
f
CS
N
am
ym
p AB
pD
12
28,800
72.193
68.841
17.476
1.437
0.277
WC-MSR
11
26,400
72.368
74.253
18.808
1.366
0.230
WC-NR
9
21,600
76.960
78.092
19.611
1.441
0.197
WC-MSR
10
24,000
69.758
83.170
20.220
2.789
0.154
WCPC-NR
10
24,000
47.813
54.949
13.579
2.518
0.421
WCPC-MSR
12
28,800
52.429
67.187
16.277
3.165
0.306
Total
64
153,600
64.998
70.830
17.600
2.129
0.266
Averages are based on minutes [5, 25] of the game phase.
263
Appendix B Statistical Analyses
F IGURE B.1: Period averages of wholesale (dashed) and retail (solid) market prices. Graphs are
plotted based on medians of 50 ticks.
300 600 900 1200 1500 1800
0
300 600 900 1200 1500 1800
300 600 900 1200 1500 1800
0
300 600 900 1200 1500 1800
WCPC-MSR
0
Time in seconds
264
300 600 900 1200 1500 1800
WC-MSR
0 20 40 60 80 100
WCPC-NR
0
0
0 20 40 60 80 100
0 20 40 60 80 100
WC-NR
0 20 40 60 80 100
Average market price
0
WM-MSR
0 20 40 60 80 100
0 20 40 60 80 100
WM-NR
300 600 900 1200 1500 1800
B.2 Wholesale competition and Open Access regulation
Linear regression of market performance indicators
X jt = b 0 + b Period · t + bWC · WC
+ bWCPC · WCPC
+ b MSR · MSR
+ bWCxMSR · WC · MSR
+ bWCPCxMSR · WCPC · MSR
+ e j,t
TABLE B.10: Linear regression of wholesale market prices am , retail market prices ym , integrated firms’ average profits p AB , retailer’s profits p D and consumer surplus
f
CS.
Baseline: Wholesale Monopoly under No Regulation (WM-NR).
Covariate
(1)
am
(2)
ym
(3)
p AB
(4)
pD
(5)
f
CS
Wholesale
Competition (WC)
4.768
(9.504)
9.251
(7.990)
2.135
(2.075)
0.004
(0.587)
0.080
(0.074)
WC w/ Price
Competition (WC)
-24.380⇤⇤⇤
(9.130)
13.892⇤
(7.645)
3.897⇤
(1.973)
1.082⇤⇤⇤
(0.406)
0.144⇤
(0.073)
Margin Squeeze
Regulation (MSR)
0.175
(8.100)
5.412
(6.624)
1.333
(1.692)
0.071
(0.383)
0.047
(0.062)
WC x MSR
-7.377
(12.972)
0.334
(10.884)
0.724
(2.800)
1.419⇤
(0.755)
0.004
(0.102)
WCPC x MSR
4.441
(12.501)
6.826
(10.339)
1.365
(2.668)
0.718
(0.573)
0.067
(0.099)
Period
0.004⇤⇤
(0.001)
0.003⇤⇤⇤
(0.001)
0.001⇤⇤⇤
(< 0.001)
Constant
65.707⇤⇤⇤
(5.886)
62.550⇤⇤⇤
(4.647)
15.929⇤⇤⇤
(1.175)
1.539⇤⇤⇤
(0.344)
0.336⇤⇤⇤
(0.043)
153,600
153,600
153,600
153,600
Observations
153,600
> 0.001
(< 0.001)
> 0.001⇤⇤⇤
(< 0.001)
Clustered standard errors (by market cohort) in parentheses.
⇤ p < 0.10, ⇤⇤ p < 0.05, ⇤⇤⇤ p < 0.01
265
Appendix B Statistical Analyses
B.3 Number effects and tacit collusion in oligopolies
Meta-regressions of tacit collusion in oligopoly experiments that vary the
number of competing firms
The use of multilevel regression models in meta-analyses has a shortcoming: The implicit weights associated to each observation, i.e., each treatment in a study, are of equal
magnitude. However, each of these values stems from an experiment designed to predict a true effect. In other words, the averages of the degree of tacit collusion in each
treatment of a study (i.e., the sample means) used in the analysis here are estimators
of the true degree of tacit collusion (i.e., the population mean) in duopolies, triopolies,
and quadropolies, respectively. Consequently, one might argue that the standard error
of each sample mean should be considered as an indication of a sample mean’s reliability. Meta-regression, a method vastly used in medical research (see, e.g., Higgins
and Thompson, 2002), does exactly this by using the within-treatment standard errors
as the standard deviations of the normal error terms in the model. More specifically,
a random-effects meta-regression model is estimated which allows for between-study
variance not explained by the covariates, i.e., the dummies for the number of firms.1
This yields a weighted regression in which the inverse of the sum of the estimated
between-study variance and the estimates’ within-treatment variances are the individual weights associated to each treatment.
Table B.11 depicts the estimates of meta-regression models with the same dependent
and independent variables as in the multilevel mixed-effects regressions. Note that
the number of observations in the meta-regressions is lower than in the corresponding mixed-effects models, because standard errors of treatment averages could not be
1 The
estimates reported in Table B.11 are derived with the metareg command of the statistical software
package Stata in its version 12. See Harbord and Higgins (2008) for further information on the command.
266
B.3 Number effects and tacit collusion in oligopolies
TABLE B.11: Meta-regression of tacit collusion on number of competitors and competition
model on the basis of most comparable treatments.
Baseline: Triopoly & Bertrand competition.
(1)
j Nash
(2)
jWalras
Duopoly
0.269⇤⇤⇤
(0.077)
0.308⇤⇤⇤
(0.073)
Quadropoly
0.026
(0.082)
0.083
(0.082)
Cournot
0.182⇤⇤
(0.076)
0.320⇤⇤⇤
(0.067)
Constant
0.023
(0.063)
0.064
(0.064)
21
21
Covariate
Observations
Baseline: Bertrand triopoly. Standard errors in parentheses.
⇤ p < 0.10, ⇤⇤ p < 0.05, ⇤⇤⇤ p < 0.01
gathered from all studies.2 For this reason, the treatment Easy in Bosch-Domènech and
Vriend (2003) cannot be considered in the meta-regressions.
Asymmetric market power and tacit collusion
Above and beyond number effects, asymmetry is expected to hinder coordination
among firms as most of the economic literature suggests that symmetry is a driver of
the ability to collude (see, e.g., Ivaldi et al., 2003; Fonseca and Normann, 2008). The
hypothesis is thus that the degrees of tacit collusion based on Nash as well as Walrasian prices and profits are significantly lower in markets with asymmetric firms than
in markets with symmetric firms, everything else being equal.
Although there is only a single difference in the parametrization between each asymmetry treatment and its symmetry counterpart, the necessary adjustment of D according to
2 The
standard errors of the degree
are derived
with the followr of tacit⇣ collusion estimates
r
⌘2
⇣
⌘2
E
E
N
N
1
x x
x x
1
x x
p1
ing relationship: SE j Ex = p1
=
=
Â
Â
JPM
E
JPM
E
JPM
E
i
=
1
i
=
1
N
1
N
1
x
x
x
x
x
x
N
N
q
2
N
1
1
p1
x ) = x JPM1 x E SE( x ) with N as the number of independent observations
N 1 Â i =1 ( x
x JPM x E N
for the corresponding treatment.
267
Appendix B Statistical Analyses
TABLE B.12: Multilevel mixed-effects linear regressions of tacit collusion on (a)symmetry of
firms in triopolies.
Baseline: Symmetric Bertrand Triopoly (B3).
(1)
j Nash
(2)
jWalras
Asymmetry
0.185
(0.149)
0.123
(0.099)
Period
0.004⇤⇤⇤
(0.001)
0.003⇤⇤⇤
(0.001)
Constant
0.685⇤⇤⇤
(0.106)
0.790⇤⇤⇤
(0.070)
Covariate
Cohorts
Observations
24
1,440
24
1,440
Baseline: Symmetric Bertrand triopoly. Standard errors in parentheses.
⇤ p < 0.10, ⇤⇤ p < 0.05, ⇤⇤⇤ p < 0.01
the number of firms in the market impedes a simultaneous analysis of asymmetry and
the number of firms. In other words, the treatment dummy between asymmetric triopolies and quadropolies is not the same as the one between symmetric triopolies and
quadropolies. Therefore, in order to assess the effect of the specific type of asymmetry
implemented here, i.e., providing a single firm with a 50% higher Nash profit than its
competitors, requires separate investigations of triopolies and quadropolies.
Tables B.12 and B.13 depict estimates of multilevel mixed-effects linear regression models of tacit collusion on symmetry and asymmetry of firms whilst controlling for heteroscedasticity via a random intercept for the cohort and a random slope for the time
trend, i.e.,
E
jk,t
= b0 + x k
+ b Asymmetry · Asymmetry
+ ( b Period + b Period,k ) · t
+ ek,t ,
268
B.3 Number effects and tacit collusion in oligopolies
TABLE B.13: Multilevel mixed-effects linear regressions of tacit collusion on (a)symmetry of
firms in quadropolies.
Baseline: Symmetric Bertrand Quadropoly (B4).
(1)
j Nash
(2)
jWalras
Asymmetry
0.179⇤⇤
(0.075)
0.135⇤⇤
(0.056)
Period
0.001
(0.001)
0.001
(0.001)
Constant
0.456⇤⇤⇤
(0.054)
0.592⇤⇤⇤
(0.040)
Covariate
Cohorts
Observations
33
1,980
33
1,980
Baseline: Symmetric Bertrand quadropoly. Standard errors in parentheses.
⇤ p < 0.10, ⇤⇤ p < 0.05, ⇤⇤⇤ p < 0.01
in triopolies and quadropolies, respectively. For maximum comparability, only the
Bertrand treatments are included in the analysis for which there is data with both symmetric and asymmetric firms. In line with the hypothesis, the degree of tacit collusion is 18 pp (13 pp) lower in quadropolies with asymmetric compared to symmetric
firms relative to the Nash (Walrasian) equilibrium. The degree of tacit collusion is not
significantly lower with asymmetry in triopolies, although the average effect size of
asymmetry is similar to the effect found for quadropolies. Given the lower number of
observations this points to a lack of statistical power in the analysis of triopolies.
Prominent theories of fairness and equity in the behavioral sciences (e.g., Fehr and
Schmidt, 1999; Bolton and Ockenfels, 2000) suggest that cooperation is harder to sustain in asymmetric than in symmetric games, which is in line with the finding here. In
an effort to assess whether social preferences in fact account for the effect of asymmetry
on tacit collusion, subjects’ social value orientation is measured using the Murphy et al.
(2011) questionnaire, which is filled out by every participant directly after the oligopoly
experiment at the end of a session, with the exception of one session with twelve participants in May 2016. Remember that incumbent firms are provided with higher market
power than entrants in the asymmetry treatments. A comparison of the continuous so-
269
Appendix B Statistical Analyses
cial value orientation index reveals no differences between incumbents and entrants or
subjects in triopolies and quadropolies. Among the four idealized social orientations
of altruistic, prosocial, individualistic, and competitive behavior, the average participant is on the verge of prosocial and individualistic behavior. This finding is further
corroborated in a categorical analysis which matches subjects to a single category. According to the classification, 43% of subjects are prosocials, 46% are individualists, and
none are altruists or of competitive type—the remaining 11% cannot be assigned due to
incomplete questionnaires. Again social orientations are not significantly different between subjects acting as firms of different types or participating in different treatments.
Furthermore, social value orientations are neither correlated with the degree of tacit
collusion connected to price decisions nor with subjects’ total profit in the experiment.
In sum, these findings suggest that social orientations cannot explain why asymmetric
firms collude less than symmetric firms.
The experimental evidence for quadropolies suggests that asymmetry fosters competition between firms in a market considerably and significantly. Put into context, the effect size of implementing asymmetry in a quadropoly by increasing the market power
of a single firm is comparable to the number effect on tacit collusion between markets
with two and three firms as well as between markets with three and four firms.
270
B.3 Number effects and tacit collusion in oligopolies
Supplementary tables and figures
TABLE B.14: Multilevel mixed-effects linear regressions of tacit collusion on number of competitors and competition model on the basis of all treatments.
Baseline: Triopoly & Bertrand competition.
(1)
j Nash
(2)
jWalras
Duopoly
0.208⇤⇤⇤
(0.037)
0.233⇤⇤⇤
(0.030)
Quadropoly
0.041
(0.046)
0.009
(0.037)
Cournot
0.249⇤⇤⇤
(0.074)
0.316⇤⇤⇤
(0.047)
Constant
0.077
(0.056)
0.138⇤⇤⇤
(0.047)
9
10
23
9
10
23
Covariate
Groups (s)
Groups (m)
Observations
Baseline: Bertrand triopoly. Standard errors in parentheses.
⇤ p < 0.10, ⇤⇤ p < 0.05, ⇤⇤⇤ p < 0.01
TABLE B.15: Inter-study average degrees of tacit collusion and one-tailed matched-samples
Wilcoxon signed-rank tests on the basis of all treatments.
Studies
j Nash
jWalras
2 vs. 3
Duopoly
Triopoly
p value
7
7
7
0.110
0.079
0.009
0.480
0.260
0.009
2 vs. 4
Duopoly
Quadropoly
p value
6
6
6
0.302
0.025
0.014
0.452
0.204
0.014
3 vs. 4
Triopoly
Quadropoly
p value
3
3
3
0.035
0.049
0.946
0.196
0.174
0.500
271
Appendix B Statistical Analyses
F IGURE B.2: Average degrees of tacit collusion jWalras over periods across treatments.
Cournot Treatments
0
.2
.4
.6
.8
1
Bertrand Treatments
0
10
20
30
40
Duopoly
272
50
60
0
Triopoly
10
20
30
Quadropoly
40
50
60
Appendix C
Experimental Instructions
For each experiment, the instructions of only one exemplary treatment are reported in
the following. The instructions for the other treatments are identical except for text passages that describe the specifics of the particular treatment. Note that the experimental instructions are translated from German and are only translations for information;
they are not intended to be used in the lab. The instructions in the original language
are carefully polished in grammar, style, comprehensibility, and avoidance of strategic
guidance.
273
Appendix C Experimental Instructions
C.1 Oligopoly competition in continuous time
The following experimental instructions have been used for the RB3 treatment.
Preliminary remarks
Welcome to the experiment and thank you very much for your participation. In this
experiment you can earn an amount of money that depends on your decisions and the
decisions of the other participants. Please address the person in charge of the experiment in case of questions. Please do not talk to the other participants during the entire
experiment. Throughout the experiment we will use the currency Euro and its subunit
cent. The Euro that you will have earned by the end of the experiment will be paid to
you in cash.
Experimental structure
There are three firms competing with each other:
Firm A
Firm B
Firm C
Each firm is represented by one participant of the experiment. Throughout the experiment the same firms compete with each other. Which firm you represent is randomly
chosen at the beginning of the experiment. Each firm offers a good that is demanded
by consumers. There are no cost for producing these goods. You choose the price at
which you want to sell your good. The quantity demanded of your good is determined
by your price. Thereby, the following holds:
The higher your price, the lower the quantity demanded of your good. Thereby, the
quantity demanded of your good can fall to zero.
The higher a price of the other firms, the higher the quantity demanded of your good.
Thereby, only prices of firms selling a positive quantity are relevant.
Your profit is calculated by multiplying your price with the quantity demanded of your
good.
274
C.1 Oligopoly competition in continuous time
Experimental procedure
The experiment is composed of two stages. At stage one you choose your initial price.
Before making the final decision, you can test how a price combination affects the quantities and profits of all firms. After all firms made their final decision by pressing the
button “Finalize decision”, the second stage of the experiment begins. The second stage
lasts exactly 30 minutes. During this time all decisions are made in real-time and without any interruptions. Your price decision is valid until you change your price. Every
decision of a firm is immediately visible for all other firms.
Software display
F IGURE C.1: Display of the experimental software.
Figure C.1 depicts the display of the experimental software. To distinguish the firms,
their information is colored as follows:
Firm A: BLUE
Firm B: GREEN
275
Appendix C Experimental Instructions
Firm C: ORANGE
In the following, the individual parts of the display will be explained bottom up.
Decision and testing environment
On the left side you can set your price by using the slider of the firm. Please be aware
that you can use all sliders during the first stage of the experiment and only the slider
of your own firm during the second stage. During the second stage the sliders show
the current prices of the other firms.
Prices
On the left side the history of all firms’ prices as well as the average price is visualized.
On the right side the current prices are displayed.
Quantities
On the left side the history of all firms’ quantities is visualized. On the right side the
current quantities are displayed.
Profits
On the left side the history of all firms’ profits is visualized. On the right side the current
profits are displayed. Please be aware that current profits are scaled to the profit you
would earn if the current combination of all prices would be held for 30 seconds. As
soon as one firm changes its price, the profits are recalculated. Your current profit is
added to your account proportionally several times per second.
Status of the experiment
On the left side it is displayed which firm you represent. Figure C.1 shows this for firm
A as an example. During the first stage there is a button “Finalize decision” in the middle of the display. Please press it when you are ready to finalize your decision. On the
right side your current account balance and the remaining duration of the experiment
is displayed. Your current account balance is the sum of all realized profits.
276
C.1 Oligopoly competition in continuous time
Concluding remarks
Before the experiment starts, you will be asked some comprehension questions on the
screen with regard to the understanding of the rules and the course of the experiment.
Please enter the respective answers into your computer. Afterwards, the experiment
will start automatically and it will be displayed which firm you represent.
In case of any questions, please remain seated and give the person in charge of the
experiment a hand signal. Please wait until the person in charge of the experiment has
arrived at your seat. Talk as quietly as possible when asking your question. Please
remain seated after the end of the experiment as well and wait for further instructions
from the person in charge of the experiment.
277
Appendix C Experimental Instructions
C.2 Wholesale competition and Open Access regulation
The following experimental instructions have been used for the WCPC
MSR treatment.
Preliminary remarks
Welcome to the experiment and thank you very much for your participation.In this experiment you can earn an amount of money that depends on your decisions and the
decisions of the other participants. Please address the person in charge of the experiment in case of questions. Please do not talk to the other participants during the entire
experiment. Throughout the experiment we will use the currency Euro and its subunits
cent. At the beginning of the experiment your account balance is EUR 5.00. At the end
of the experiment, the final account balance will be paid to you in cash.
During the experiment you represent a firm which is selling a good to consumers. Next
to you, there are two other firms which are competing with you. All your decisions
are made in real time, thus, they are immediately effective and visible to all other
firms. Over the entire time horizon of the experiment, you play together with the same
firms.
Experimental structure
There are three firms:
Firm A
Firm B
Firm C
Firm A and Firm B are represented by participants of the experiment. Firm C acts
computerized. Which firm you represent is randomly chosen at the beginning of the
experiment and does not change over the entire experiment. Furthermore there are two
markets:
Wholesale market
Retail market
Figure C.2 visualizes the structure of the experiment. Each of the three firms offers
a retail product on the retail market and chooses its retail price. In order to produce
the retail product each firm needs a wholesale product. Only Firm A and Firm B offer
the wholesale product in the wholesale market and choose their respective wholesale
278
C.2 Wholesale competition and Open Access regulation
prices. Firm C has to buy the wholesale product from one of the two other firms in
order to be able to offer its retail product.
F IGURE C.2: Structure of the experiment.
Wholesalemarket
FirmB
FirmA
Wholesale
good
Wholesaleprice
Retailmarket
Wholesaleprice
Wholesale
good
FirmC
Retailgood
Retailprice
Retailgood
Retailprice
Retailgood
Retailprice
Consumers
Wholesale Market
The wholesale products of Firm A and Firm B are equal. Thereby, the following holds:
Firm C chooses automatically the cheaper wholesale product to satisfy its demand.
If Firm A and Firm B offer the identical wholesale price, Firm C chooses the wholesale
product from the firm which had previously offered the lower price.
If Firm A and Firm B offer the identical wholesale price at the beginning, Firm C
chooses randomly from which firm it purchases the wholesale product.
There are no handling costs for the wholesale product. The prices of the wholesale
products range from 0 to 100.
Retail market
The retail products differ between firms. The demand of your retail product depends
on your retail price and the retail prices of the other firms. Thereby, the following holds
under the assumption that the other retail prices remain unchanged:
If you increase your retail price, the demand of your retail product decreases.
If one of the other firms increases its retail price, the demand of your retail products
increases.
If all firms increase their retail price, the total demand of all retail products decreases.
279
Appendix C Experimental Instructions
If your retail price is located below the average of all three retail prices, the demand of
your retail product increases. If your retail price is located above the average of all three
retail prices, the demand of your retail product decreases. The extent of the deviation of
your retail price from the average of all three retail prices determines the magnitude of
this effect. If your retail price is above the average of all three retail prices, the demand
of your retail product may fall to zero. Firm C chooses its profit-maximizing retail price
in reaction to the effective wholesale price and the retail prices chosen by Firm A and
Firm B.
There are no handling costs for the retail product. The prices of the retail products
range from 0 to 100.
Profits
The profits of the three firms depend on the retail and wholesale prices. The calculations
for the profits of Firm A and Firm B depend on Firm C’s decision which firm to choose
as its wholesale provider.
If Firm C chooses to purchase its wholesale product from Firm A, the following holds
for the profits of each firm:
Pro f it A = Retail Price A · Demand A + Wholesale Price A · DemandC
Pro f it B = Retail Price B · Demand B
Pro f itC = ( Retail PriceC
Wholesale price A ) · DemandC
If Firm C chooses to purchase its wholesale product from Firm B, the following holds
for the profits of each firm:
Pro f it A = Retail Price A · Demand A
Pro f it B = Retail Price B · Demand B + Wholesale Price B · DemandC
Pro f itC = ( Retail PriceC
Wholesale Price B ) · DemandC
Experimental procedure
The experiment is composed of two stages. At the first stage, as Firm A or Firm B, you
choose your initial retail price and your initial wholesale price. Before making your
final decision, you can test how a price combination affects the profits of all three firms.
This does not influence your account balance. After all firms have made their initial
price decision and have confirmed their decisions with a click on “apply initial prices”,
the second stage of the experiment starts.
280
C.2 Wholesale competition and Open Access regulation
The second stage lasts exactly 30 minutes. During this period of time, all decisions are
made in real time and without any interruptions. Your price decision remains effective
until you change your price. Note that subsequent to a change of your wholesale price,
the price cannot be changed again for the next 30 seconds. Furthermore, please be
aware that your wholesale price can not be located above your retail price.
Software display
F IGURE C.3: Display of the experimental software.
Figure C.3 depicts the display of the exeriment software. In order to distinguish the
firms, their labels are colored as follows:
Firm A: BLUE
Firm B: GREEN
Firm C: ORANGE
In the following, the individual sections of the display will be explained from the bottom up:
281
Appendix C Experimental Instructions
Experimental progress
On the left-hand side, it is denoted whether you represent Firm A or Firm B. The figure
illustrates this exemplarily for Firm A. On the right-hand side, your current account
balance as well as the remaining duration of the experiment is displayed. Your current
account balance consists of the initial balance of EUR 5.00 and the additionally earned
profits during the experiment.
Current profits and profit history
On the right-hand side, the current profits of all firms are displayed. Note that current
profits are scaled to the profit you would earn, if the current combination of all prices
would be held over the entire 30 minutes of the experiment. As soon as one of the
prices changes, the current profits are recalculated. On the left-hand side, the history of
the current profits is displayed.
Current prices and price history
On the right-hand side, the current prices of all three firms are displayed. The effective
wholesale price is always the lower wholesale price of both wholesale prices. On the
left-hand side, the history of your retail price, the average retail price of all three firms
and the effective wholesale price is displayed.
Wholesale prices and current profits in the wholesale market
On the left-hand side, Firm A and Firm B choose their wholesale prices. Be aware that
Firm C offers no wholesale product and thus cannot choose a wholesale price. The
wholesale price can be set with the corresponding slider by using the mouse or the
arrow keys on the keyboard. Note that you can move all sliders at the first stage of the
experiment and only the slider of your firm at the second stage of the experiment. The
sliders of the other firms show their current wholesale prices. On the right-hand side
the current profits in the wholesale market are displayed. Furthermore it is displayed
which firms sells its wholesale product to Firm C. Note that subsequent to a change
of your wholesale price, the price cannot be changed again for the next 30 seconds.
Furthermore, please be aware that your wholesale price can not be located above your
retail price.
282
C.2 Wholesale competition and Open Access regulation
Retail prices and current profits in the retail market
On the left-hand side, all of the three firms choose their retail price. The retail price
can be set with the corresponding slider by using the mouse or the arrow keys on the
keyboard. Note that you can move all sliders at the first stage of the experiment and
only the slider of your firm at the second stage of the experiment. The sliders of the
other firms show their current retail prices. On the right-hand side, the current profits
in the retail market are displayed. Note that the displayed current profit of Firm C
already includes the costs for the wholesale product.
Concluding remarks
Before the experiment starts, you will be asked a set of comprehension questions, displayed on the computer screen, that cover the rules and the procedure of the experiment. Please enter the respective answers. Thereupon, the experiment will start automatically and it is displayed which firm you represent.
In case of any questions during the experiment, please remain seated and inform the
person in charge of the experiment by the means of a hand gesture. Please wait until the
person in charge of the experiment has arrived at your seat. Talk as quietly as possible
when asking your question. Please remain seated after the end of the experiment and
wait for further instructions from the person in charge of the experiment.
283
Appendix C Experimental Instructions
C.3 Number effects and tacit collusion in oligopolies
The following experimental instructions have been used for the B4A treatment.
Preliminary remarks
Welcome to the experiment and thank you very much for your participation. In this
experiment you can earn an amount of money that depends on your decisions and the
decisions of the other participants. Please address the person in charge of the experiment in case of questions. Please do not talk to the other participants during the entire
experiment. Throughout the experiment we will use the currency Euro and its subunit
cent. The Euro that you will have earned by the end of the experiment will be paid to
you in cash.
Experimental structure
There are four firms competing with each other:
Firm A
Firm B
Firm C
Firm D
Each firm is represented by one participant of the experiment. Throughout the experiment the same firms compete with each other. Which firm you represent is randomly
chosen at the beginning of the experiment. Each firm offers a good that is demanded
by consumers. There are no cost for producing these goods. You choose the price at
which you want to sell your good. The quantity demanded of your good is determined
by your price. Thereby, the following holds:
The higher your price, the lower the quantity demanded of your good. Thereby, the
quantity demanded of your good can fall to zero.
The higher a price of the other firms, the higher the quantity demanded of your good.
Thereby, only prices of firms selling a positive quantity are relevant.
Please be aware that the quantity demanded of Firm A differs from the quantity demanded of any other firm, everything else being equal. If all firms choose the same
price, the quantity demanded of Firm A is greater than the quantity demanded of any
other single firm. Additionally, the following holds:
284
C.3 Number effects and tacit collusion in oligopolies
If Firm A raises its price, the quantity demanded of Firm A decreases less than the
quantity of any other firm if it raises its price.
If another firm raises its price, the quantity demanded of Firm A increases more than
the quantity of any other firm.
Your profit is calculated by multiplying your price with the quantity demanded of your
good.
Experimental procedure
The experiment lasts 60 periods. In each period you chose your price. Before making
the final decision, you can test how a price combination affects the quantities and profits
of all firms. After all firms made their final decision by pressing the button “Finalize
decision”, quantities and profits are calculated and the next period begins.
Software display
F IGURE C.4: Display of the experimental software.
285
Appendix C Experimental Instructions
Figure C.4 depicts the display of the experimental software. To distinguish the firms,
their information is colored as follows:
Firm A: BLUE
Firm B: GREEN
Firm C: ORANGE
Firm D: PURPLE
In the following, the individual parts of the display will be explained bottom up.
Decision and testing environment
On the left side you can set your price by using the slider of your firm. At the beginning
of a period the sliders show the prices of the firms of the previous period. Before making the final decision, you can test the consequences of your price decision by adjusting
the sliders of the other firms to your expectations. As soon as you release a slider, the
quantities and profits that would result in the next period if the currently set prices in
the testing environment get chosen are displayed on the right side of the screen above
the sliders. On the right side you can reset the sliders to the prices of the last period by
pressing the button “Show last period results”.
Prices
On the left side the history of all firms’ prices as well as the average price is visualized.
On the right side the currently set prices in the testing environment are displayed.
Quantities
On the left side the history of all firms’ quantities is visualized. On the right side the
quantities that would result in the next period if the currently set prices in the testing
environment get chosen are displayed.
Profits
On the left side the history of all firms’ profits is visualized. On the right side the profits
that would result in the next period if the currently set prices in the testing environment
get chosen are displayed.
286
C.3 Number effects and tacit collusion in oligopolies
Status of the experiment
On the left side it is displayed which firm you represent. Figure C.4 shows this for firm
A as an example. In the middle of the display is the button “Finalize decision”. Please
press it when you are ready to finalize your decision. On the right side your current
account balance and the current period of the experiment is displayed. Your current
account balance is the sum of all realised profits.
Concluding remarks
Before the experiment starts, you will be asked some comprehension questions on the
screen with regard to the understanding of the rules and the course of the experiment.
Please enter the respective answers into your computer. Afterwards, the experiment
will start automatically and it will be displayed which firm you represent.
In case of any questions, please remain seated and give the person in charge of the
experiment a hand signal. Please wait until the person in charge of the experiment has
arrived at your seat. Talk as quietly as possible when asking your question. Please
remain seated after the end of the experiment as well and wait for further instructions
from the person in charge of the experiment.
287
References
5G PPP (2015).
ship:
5G vision. The 5G infrastructure public private partner-
the next generation of communication networks and services.
Avail-
able at https://5g-ppp.eu/wp-content/uploads/2015/02/5G-VisionBrochure-v1.pdf. Accessed May 05, 2016.
Abbink, K. and Brandts, J. (2005). Price competition under cost uncertainty: A laboratory analysis. Economic Inquiry, 43(3):636–648.
Abbink, K. and Brandts, J. (2008). 24. Pricing in Bertrand competition with increasing
marginal costs. Games and Economic Behavior, 63(1):1–31.
Acquisti, A., Taylor, C. R., and Wagman, L. (2016). The economics of privacy. Journal of
Economic Literature, 54(2):442–492.
Acquisti, A. and Varian, H. R. (2005). Conditioning prices on purchase history. Marketing Science, 24(3):367–381.
Aghion, P., Bechtold, S., Cassar, L., and Herz, H. (2014). The causal effects of competition on innovation: Experimental evidence. Working Paper. Available at http:
//www.nber.org/papers/w19987.
Aghion, P., Bloom, N., Blundell, R., Griffith, R., and Howitt, P. (2005). Competition
and innovation: An inverted-U relationship.
The Quarterly Journal of Economics,
120(2):701–728.
289
References
Aguirre, E., Mahr, D., Grewal, D., de Ruyter, K., and Wetzels, M. (2015). Unraveling
the personalization paradox: The effect of information collection and trust-building
strategies on online advertisement effectiveness. Journal of Retailing, 91(1):34–49.
Alger, D. (1987). Laboratory tests of equilibrium predictions with disequilibrium data.
The Review of Economic Studies, 54(1):105–145.
Ambrus, A., Calvano, E., and Reisinger, M. (2016). Either or both competition: A “twosided” theory of advertising with overlapping viewerships. American Economic Journal: Microeconomics. Article in advance.
Anand, B. and Shachar, R. (2009). Targeted advertising as a signal. Quantitative Marketing and Economics, 7(3):237–266.
Anderson, S. A. (2012). Advertising on the internet. In Peitz, M. and Waldfogel, J.,
editors, The Oxford Handbook of the Digital Economy. Oxford University Press, New
York, NY.
Anderson, S. P. and Coate, S. (2005). Market provision of broadcasting: A welfare
analysis. The Review of Economic Studies, 72(4):947–972.
Anderson, S. P. and De Palma, A. (2013). Shouting to be heard in advertising. Management Science, 59(7):1545–1556.
Angrist, J. D. and Pischke, J.-S. (2010). The credibility revolution in empirical economics: How better research design is taking the con out of econometrics. Journal
of Economic Perspectives, 24(2):3–30.
Animesh, A., Ramachandran, V., and Viswanathan, S. (2010). Research note-Quality
uncertainty and the performance of online sponsored search markets: An empirical
investigation. Information Systems Research, 21(1):190–201.
Ansari, A. and Mela, C. F. (2003). E-Customization. Journal of Marketing Research,
40(2):131–145.
290
References
Arakali, H. (2015). Facebook now de facto internet gateway for millions of Indians
through Internet.org.
Available at http://www.ibtimes.com/facebook-
now-de-facto-internet-gateway-millions-indians-throughinternetorg-1811704. Accessed March 02, 2016.
Argenton, C. and Prüfer, J. (2012). Search engine competition with network externalities. Journal of Competition Law & Economics, 8(1):73–105.
Armstrong, M. (2002). The theory of access pricing and interconnection. In Cave, M.,
Vogelsang, I., and Majumdar, S., editors, Handbook of Telecommunications Economics,
volume 1. North-Holland, Amsterdam, Netherlands.
Armstrong, M. (2006). Competition in two-sided markets. The RAND Journal of Economics, 37(3):668–691.
Armstrong, M., Doyle, C., and Vickers, J. (1996). The access pricing problem: A synthesis. The Journal of Industrial Economics, 44(2):131–150.
Armstrong, M. and Vickers, J. (1998). The access pricing problem with deregulation: A
note. The Journal of Industrial Economics, 46(1):115–121.
Arthur, W. B. (2006). Out-of-equilibrium economics and agent-based modeling. In
Tesfatsion, L. and Judd, K. L., editors, Handbook of Computational Economics, volume 2,
pages 1551–1564. North-Holland, Amsterdam, Netherlands.
Asdemir, K., Kumar, N., and Jacob, V. S. (2012). Pricing models for online advertising:
CPM vs. CPC. Information Systems Research, 23(3-part-1):804–822.
A.T. Kearney (2016).
Competition and innovation.
Available at http://www.
gsma.com/publicpolicy/wp-content/uploads/2016/05/GSMA_Theinternet-Value-Chain_WEB.pdf. Accessed May 20, 2016.
Athey, S., Calvano, E., and Gans, J. S. (2014). The impact of the internet on advertising markets for news media. Working Paper. Available at http://ssrn.com/
291
References
abstract=2180851.
Athey, S. and Gans, J. S. (2010). The impact of targeting technology on advertising
markets and media competition. American Economic Review, 100(2):608.
Atiyas, I., Doganoglu, T., and Inceoglu, F. (2015). Upstream competition and complexity of contracts. Working Paper. Available at https://editorialexpress.com/
cgi-bin/conference/download.cgi?db_name=EARIE42&paper_id=501.
Auf’mkolk, H. (2012). From regulatory tool to competition law rule: The case of margin
squeeze under EU competition law. Journal of European Competition Law & Practice,
3(2):149–162.
Avison, D. and Malaurent, J. (2014). Is theory king?: questioning the theory fetish in
information systems. Journal of Information Technology, 29(4):327–336.
Bacache, M., Bourreau, M., and Gaudin, G. (2014). Dynamic entry and investment in
new infrastructures: Empirical evidence from the fixed broadband industry. Review
of Industrial Organization, 44(2):179–209.
Backhouse, R. E. (2012). The rise and fall of Popper and Lakatos in economics. In Mäki,
U., editor, Philosophy of Economics, volume 13 of Handbook of the Philosophy of Science,
pages 25–48. Elsevier, Amsterdam, Netherlands.
Baker, M. (2016). Is there a reproducibility crisis? Nature, 533(7604):452–454.
Bakos, Y. and Brynjolfsson, E. (1999). Bundling information goods: Pricing, profits, and
efficiency. Management Science, 45(12):1613–1630.
Ball, S. and Cech, P.-A. (1996). Subject pool choice and treatment effects in economic
laboratory research. In Isaac, R. and Norton, D., editors, Research in Experimental
Economics, volume 6, pages 239–292. JAI Press, Greenwich, CT.
292
References
Ballon, P. and Van Heesvelde, E. (2011). ICT platforms and regulatory concerns in
Europe. Telecommunications Policy, 35(8):702–714.
Balmer, R. E. (2013). Entry and competition in local newspaper retail markets - When
two are enough.
Working Paper. Available at http://ssrn.com/abstract=
2369040.
Banerjee, A. and Dippon, C. M. (2009). Voluntary relationships among mobile network
operators and mobile virtual network operators: An economic explanation. Information Economics and Policy, 21(1):72–84.
Barnett, E. (2010). Facebook: the gateway onto the internet?
Available at http:
//www.telegraph.co.uk/technology/facebook/8037491/Facebookthe-gateway-onto-the-internet.html. Accessed March 02, 2016.
Battiti, R., Cigno, R. L., Sabel, M., Orava, F., and Pehrson, B. (2005).
Wireless
LANs: From warchalking to open access networks. Mobile Networks and Applications,
10(3):275–287.
Bauer, J. M. (2010). Regulation, public policy, and investment in communications infrastructure. Telecommunications Policy, 34(1):65–79.
Bauer, J. M. (2014). Platforms, systems competition, and innovation: Reassessing the
foundations of communications policy. Telecommunications Policy, 38(8):662–673.
Bauer, J. M. and Bohlin, E. (2008). From static to dynamic regulation. Intereconomics,
43(1):38–50.
Baumol, W. J. (2001). When is inter-firm coordination beneficial? The case of innovation.
International Journal of Industrial Organization, 19(5):727–737.
Baumol, W. J., Ordover, J. A., and Willig, R. D. (1997). Parity pricing and its critics: A
necessary condition for efficiency in the provision of bottleneck services to competitors. Yale Journal on Regulation, 14:145–163.
293
References
Baye, M. and Morgan, J. (2001). Information gatekeepers on the internet and the competitiveness of homogeneous product markets. American Economic Review, 91(3):454–
474.
Beard, T. R., Kaserman, D. L., and Mayo, J. W. (2001). Regulation, vertical integration
and sabotage. The Journal of Industrial Economics, 49(3):319–333.
Begley, C. G. and Ellis, L. M. (2012). Drug development: Raise standards for preclinical
cancer research. Nature, 483(7391):531–533.
Bellemare, C., Bissonnette, L., and Kröger, S. (2014). Statistical power of within and
between-subjects designs in economic experiments. Working Paper. Available at
http://ssrn.com/abstract=2529895.
Belloc, F., Nicita, A., and Alessandra Rossi, M. (2012). Whither policy design for broadband penetration? Evidence from 30 OECD countries. Telecommunications Policy,
36(5):382–398.
Bengtsson, M. and Kock, S. (2000). “Coopetition” in business networks - to cooperate
and compete simultaneously. Industrial Marketing Management, 29(5):411–426.
Bengtsson, M. and Kock, S. (2014). Coopetition-Quo vadis? Past accomplishments and
future challenges. Industrial Marketing Management, 43(2):180–188.
BEREC (2011).
BEREC Report on “Open Access” (BoR (11) 05).
Available
at http://berec.europa.eu/eng/document_register/subject_matter/
berec/reports/?doc=212. Accessed July 16, 2013.
BEREC (2015). BEREC report on oligopoly analysis and regulation (BoR (15) 195).
Available at http://berec.europa.eu/eng/document_register/subject_
matter/berec/reports/5581-berec-report-on-oligopoly-analysisand-regulation. Accessed May 05, 2016.
294
References
Bergemann, D. and Bonatti, A. (2011). Targeting in advertising markets: implications
for offline versus online media. The RAND Journal of Economics, 42(3):417–443.
Bergin, J. and MacLeod, W. B. (1993). Continuous time repeated games. International
Economic Review, 34(1):21–37.
Berninghaus, S. K. and Ehrhart, K.-M. (1998). Time horizon and equilibrium selection
in tacit coordination games: Experimental results. Journal of Economic Behavior &
Organization, 37(2):231–248.
Berninghaus, S. K. and Ehrhart, K.-M. (2003). The power of ESS: An experimental study.
Journal of Evolutionary Economics, 13(2):161–181.
Berninghaus, S. K., Ehrhart, K.-M., and Keser, C. (1999). Continuous-time strategy
selection in linear population games. Experimental Economics, 2(1):41–57.
Berninghaus, S. K., Ehrhart, K.-M., and Ott, M. (2006). A network experiment in continuous time: The influence of link costs. Experimental Economics, 9(3):237–251.
Berninghaus, S. K., Ehrhart, K.-M., Ott, M., and Vogt, B. (2007). Evolution of networks
- an experimental analysis. Journal of Evolutionary Economics, 17(3):317–347.
Bichler, M., Gupta, A., and Ketter, W. (2010). Research commentary-Designing smart
markets. Information Systems Research, 21(4):688–699.
Bigoni, M., Casari, M., Skrzypacz, A., and Spagnolo, G. (2015). Time horizon and cooperation in continuous time. Econometrica, 83(2):587–616.
Binmore, K. (1987).
Modeling rational players: Part I.
Economics and Philosophy,
3(02):179–214.
Bleier, A. and Eisenbeiss, M. (2015). Personalized online advertising effectiveness: The
interplay of what, when, and where. Marketing Science, 34(5):669–688.
295
References
Block, C., Collins, J., and Ketter, W. (2010). Agent-based competitive simulation: Exploring future retail energy markets. In Proceedings of the 12th International Conference
on Electronic Commerce (ICEC), pages 68–77. Honolulu, HI.
Bock, O., Baetge, I., and Nicklisch, A. (2014). hroot: Hamburg registration and organization online tool. European Economic Review, 71:117–120.
Bogliolo, A. (2009). Introducing neutral access networks. In Proceedings of the 5th EuroNGI conference on Next Generation Internet networks (NGI), pages 243–248. Ria Aveiro,
Portugal.
Bolton, G. E. and Ockenfels, A. (2000). ERC: A theory of equity, reciprocity, and competition. American Economic Review, 90(1):166–193.
Boots, M. G., Rijkers, F. A., and Hobbs, B. F. (2004). Trading in the downstream European gas market: A successive oligopoly approach. The Energy Journal, 25(3):73–102.
Bortolotti, B., D‘Souza, J., Fantini, M., and Megginson, W. L. (2002). Privatization and
the sources of performance improvement in the global telecommunications industry.
Telecommunications Policy, 26(5):243–268.
Bosch-Domènech, A. and Vriend, N. J. (2003). Imitation of successful behaviour in
Cournot markets. The Economic Journal, 113(487):495–524.
Bouckaert, J. and Kort, P. M. (2014). Merger incentives and the failing firm defense. The
Journal of Industrial Economics, 62(3):436–466.
Bouckaert, J. and Verboven, F. (2004). Price squeezes in a regulatory environment. Journal of Regulatory Economics, 26(3):321–351.
Boudreau, K. (2010). Open platform strategies and innovation: Granting access vs.
devolving control. Management Science, 56(10):1849–1872.
296
References
Bourreau, M., Cambini, C., and Hoernig, S. (2012a).
Ex ante regulation and co-
investment in the transition to next generation access. Telecommunications Policy,
36(5):399–406.
Bourreau, M., Cambini, C., and Hoernig, S. (2012b). Geographic access rules and investment. Working Paper. Available at http://ssrn.com/abstract=2153445.
Bourreau, M., Cambini, C., and Hoernig, S. (2013). Cooperative investment, uncertainty
and access. Working Paper. Available at http://ssrn.com/abstract=2231867.
Bourreau, M., Cambini, C., and Hoernig, S. (2015). Geographic access markets and
investments. Information Economics and Policy, 31:13–21.
Bourreau, M., Hombert, J., Pouyet, J., and Schutz, N. (2011). Upstream competition
between vertically integrated firms. The Journal of Industrial Economics, 59(4):677–713.
Box, G. E. (1979). Robustness in the strategy of scientific model building. In Launer,
R. L. and Wilkinson, G. N., editors, Robustness in Statistics, pages 201–236. Academic
Press, New York, NY.
Brandts, J. and Guillén, P. (2007). Collusion and fights in an experiment with pricesetting firms and production in advance. The Journal of Industrial Economics, 55(3):453–
473.
Braun, M. and Moe, W. W. (2013). Online display advertising: Modeling the effects of
multiple creatives and individual impression histories. Marketing Science, 32(5):753–
767.
Breuer, J., Ranaivoson, H., Buchinger, U., and Ballon, P. (2015). Who manages the manager? Identity management and user ownership in the age of data. In 13th Annual
Conference on Privacy, Security and Trust (PST), pages 22–27. Izmir, Turkey.
Briglauer, W., Ecker, G., and Gugler, K. (2013). The impact of infrastructure and servicebased competition on the deployment of next generation access networks: Recent ev-
297
References
idence from the European member states. Information Economics and Policy, 25(3):142–
153.
Briglauer, W., Götz, G., and Schwarz, A. (2011). Margin squeeze in fixed-network
telephony markets - Competitive or anticompetitive? Review of Network Economics,
10(4):1–19.
Brito, D. and Pereira, P. (2010). Access to bottleneck inputs under oligopoly: A prisoners’ dilemma? Southern Economic Journal, 76(3):660–677.
Brito, D., Pereira, P., and Vareda, J. (2012). Does vertical separation necessarily reduce
quality discrimination and increase welfare? The BE Journal of Economic Analysis &
Policy, 12(1):1–42.
Broadband Forum (2012).
MR-257. An overview of G.993.5 Vectoring.
Available
at https://www.broadband-forum.org/marketing/download/mktgdocs/
MR-257.pdf. Accessed May 05, 2016.
Broadband Forum (2015). TR-301. Architecture and requirements for fiber to the distribution point. Available at https://www.broadband-forum.org/technical/
download/TR-301.pdf. Accessed May 05, 2016.
Brynjolfsson, E. and Hitt, L. M. (2000). Beyond computation: Information technology,
organizational transformation and business performance. Journal of Economic Perspectives, 14(4):23–48.
Brynjolfsson, E., Hu, Y., and Smith, M. D. (2003). Consumer surplus in the digital
economy: Estimating the value of increased product variety at online booksellers.
Management Science, 49(11):1580–1596.
Budzinski, O. and Köhler, K. H. (2015). Is Amazon the next Google? Working Paper.
Available at http://ssrn.com/abstract=2676079.
298
References
Bundeskartellamt (2016). Bundeskartellamt initiates proceeding against Facebook on
suspicion of having abused its market power by infringing data protection rules.
Available at http://www.bundeskartellamt.de/SharedDocs/Meldung/EN/
Pressemitteilungen/2016/02_03_2016_Facebook.html. Accessed May 05,
2016.
Bundesnetzagentur (2011).
NGA Forum report.
Available at http:
//www.bundesnetzagentur.de/SharedDocs/Downloads/EN/BNetzA/
Areas/Telecommunications/TelecomRegulation/NGAForum/2012Nov_
Endbericht_NGA_Forum_EN_pdf.pdf?__blob=publicationFile. Accessed
July 16, 2013.
Bundesnetzagentur (2013a). Bundesnetzagentur submits vectoring decision to European Commission.
Available at http://www.bundesnetzagentur.de/cln_
1932/SharedDocs/Pressemitteilungen/EN/2013/130709_VectoringEC.
html?nn=404530. Accessed July 16, 2013.
Bundesnetzagentur (2013b). NGA Forum. AG Interoperabilität. Leistungsbeschreibung
eines Ebene 2-Zugangsprodukts in Kabelnetzen. L2-BSA II - Technische Spezifikation.
Available at http://www.bundesnetzagentur.de/SharedDocs/
Downloads/DE/Sachgebiete/Telekommunikation/Unternehmen_
Institutionen/Breitband/NGA_NGN/NGA-Forum/aktuelledokumente/
L2_BSA_II_TechSpezifikation%20Kabel_v10_280517.pdf?__blob=
publicationFile&v=4. Accessed May 05, 2016.
Bundesnetzagentur (2015).
Festlegung der Bundesnetzagentur für Elektrizität,
Gas, Telekommunikation, Post und Eisenbahnen Für Massenmarktprodukte
auf der Vorleistungsebene an festen Standorten zentral bereitgestellter Zugang.
Available at http://www.bundesnetzagentur.de/DE/Service-
Funktionen/Beschlusskammern/1BK-Geschaeftszeichen-Datenbank/
BK3-GZ/2014/2014_0001bis0999/2014_100bis199/BK3-14-114/BK314-114_Ergänzung%20zur%20Festlegung_Regulierungsverfügung_
299
References
download.pdf?__blob=publicationFile&v=3. Accessed May 05, 2016.
Burlacu, A. (2015). Facebook outages explained: Here’s why it went down three times
last month.
Available at http://www.techtimes.com/articles/92119/
20151006/facebook-outages-explained-heres-why-it-went-downthree-times-last-month.htm. Accessed March 08, 2016.
Butters, G. R. (1977). Equilibrium distributions of sales and advertising prices. The
Review of Economic Studies, 44(3):465–491.
Button, K. S., Ioannidis, J. P., Mokrysz, C., Nosek, B. A., Flint, J., Robinson, E. S., and
Munafò, M. R. (2013). Power failure: why small sample size undermines the reliability of neuroscience. Nature Reviews Neuroscience, 14(5):365–376.
Cadman, R. (2010). Means not ends: Deterring discrimination through equivalence and
functional separation. Telecommunications Policy, 34(7):366–374.
Calvano, E. and Jullien, B. (2012). Issues in online advertising and competition policy:
A two sided markets perspective. In Harrington, J. E. and Katsoulacos, Y., editors,
Recent Advances in the Analysis of Competition Policy and Regulation, pages 179–197.
Edward Elgar Publishing, Cheltenham, UK.
Cambini, C. and Jiang, Y. (2009). Broadband investment and regulation: A literature
review. Telecommunications Policy, 33(10):559–574.
Camerer, C. F., Dreber, A., Forsell, E., Ho, T.-H., Huber, J., Johannesson, M., Kirchler,
M., Almenberg, J., et al. (2016). Evaluating replicability of laboratory experiments in
economics. Science, 351(6280):1433–1436.
Cardona, M., Kretschmer, T., and Strobel, T. (2013). ICT and productivity: conclusions
from the empirical literature. Information Economics and Policy, 25(3):109–125.
Carlton, D. W. (2008). Should “price squeeze” be a recognized form of anticompetitive
conduct? Journal of Competition Law & Economics, 4(2):271–278.
300
References
Cartwright, N. (2005). The vanity of rigour in economics: theoretical models and
Galilean experiments. In Fontaine, P. and Leonard, R., editors, The Experiment in
the History of Economics. Routledge, London, UK.
Cave, M. (2006a). Encouraging infrastructure competition via the ladder of investment.
Telecommunications Policy, 30(3):223–237.
Cave, M. (2006b). Six degrees of separation operational separation as a remedy in European telecommunications regulation. Communications & Strategies, 64(4):89–103.
Cave, M. and Vogelsang, I. (2003). How access pricing and entry interact. Telecommunications Policy, 27(10):717–727.
Chen, E. Y., Pei, Y., Chen, S., Tian, Y., Kotcher, R., and Tague, P. (2014). OAuth demystified for mobile application developers. In Proceedings of the 2014 ACM SIGSAC
Conference on Computer and Communications Security, pages 892–903. Scottsdale, AZ.
Chen, Y. (2001). On vertical mergers and their competitive effects. The RAND Journal of
Economics, 32(4):667–685.
Chen, Y. and Riordan, M. H. (2007). Vertical integration, exclusive dealing, and ex post
cartelization. The RAND Journal of Economics, 38(1):1–21.
Cheung, Y.-W. and Friedman, D. (2009). Speculative attacks: A laboratory study in
continuous time. Journal of International Money and Finance, 28(6):1064–1082.
Choi, J. and Yi, S. (2000). Vertical foreclosure with the choice of input specifications. The
RAND Journal of Economics, 31(4):717–743.
Choi, J. P. and Kim, B.-C. (2010). Net neutrality and investment incentives. The RAND
Journal of Economics, 41(3):446–471.
Claffy, K. and Clark, D. D. (2013). Platform models for sustainable internet regulation.
In Telecommunications Policy Research Conference (TPRC). Arlington, VA.
301
References
Coase, R. H. (1972). Durability and monopoly. The Journal of Law and Economics,
15(1):143–49.
comScore (2011). U.S. online display advertising market delivers 1.1 trillion impressions in Q1 2011. Available at http://www.comscore.com/Insights/PressReleases/2011/5/U.S.-Online-Display-Advertising-MarketDelivers-1.1-Trillion-Impressions-in-Q1-2011.
Accessed
March
02, 2016.
Crandall, R. and Sidak, G. J. (2002). Is structural separation of incumbent local exchange
carriers necessary for competition? Yale Journal on Regulation, 19:335–411.
Crandall, R. W., Eisenach, J. A., and Litan, R. E. (2010). Vertical separation of telecommunciations networks: Evidence from five countries. Federal Communications Law
Journal, 62:493–540.
Czernich, N., Falck, O., Kretschmer, T., and Woessmann, L. (2011). Broadband infrastructure and economic growth. The Economic Journal, 121(552):505–532.
Dal Bó, P. (2005). Cooperation under the shadow of the future: Experimental evidence
from infinitely repeated games. American Economic Review, 95(5):1591–1604.
D’Annunzio, A. and Russo, A. (2015). Net neutrality and internet fragmentation: The
role of online advertising. International Journal of Industrial Organization, 43:30–47.
Darai, D., Sacco, D., and Schmutzler, A. (2010). Competition and innovation: an experimental investigation. Experimental Economics, 13(4):439–460.
Dasgupta, P. (2002). Modern economics and its critics. In Mäki, U., editor, Fact and
Fiction in Economics: Models, Realism and Social Construction, pages 57–89. Cambridge
University Press, Cambridge, UK.
Davis, D. (2009). Pure numbers effects, market power, and tacit collusion in posted
offer markets. Journal of Economic Behavior & Organization, 72(1):475–488.
302
References
Davis, D., Korenok, O., and Reilly, R. (2009). Re-matching, information and sequencing
effects in posted offer markets. Experimental Economics, 12(1):65–86.
Davis, D., Korenok, O., and Reilly, R. (2010). Cooperation without coordination: signaling, types and tacit collusion in laboratory oligopolies. Experimental Economics,
13(1):45–65.
Davis, D. D. and Korenok, O. (2009). Posted offer markets in near-continuous time: An
experimental investigation. Economic Inquiry, 47(3):449–466.
De Bijl, P. W. (2005). Structural separation and access in telecommunications markets.
Journal of Network Industries, 6(2):95–114.
De Bijl, P. W. and Peitz, M. (2008). Innovation, convergence and the role of regulation
in the Netherlands and beyond. Telecommunications Policy, 32(11):744–754.
Deck, C. and Nikiforakis, N. (2012). Perfect and imperfect real-time monitoring in a
minimum-effort game. Experimental Economics, 15(1):71–88.
Deck, C. A. and Wilson, B. J. (2002). The effectiveness of low price matching in mitigating the competitive pressure of low friction electronic markets. Electronic Commerce
Research, 2(4):385–398.
Deck, C. A. and Wilson, B. J. (2003). Automated pricing rules in electronic posted offer
markets. Economic Inquiry, 41(2):208–223.
Deck, C. A. and Wilson, B. J. (2008). Experimental gasoline markets. Journal of Economic
Behavior & Organization, 67(1):134–149.
DeGraba, P. (1990). Input market price discrimination and the choice of technology.
American Economic Review, 80(5):1246–1253.
DeGraba, P. (2003). A bottleneck input supplier’s opportunity cost of competing downstream. Journal of Regulatory Economics, 23(3):287–297.
303
References
Deutsche Telekom (2011).
Open access for new networks.
Available at http:
//www.telekom.com/media/media-kits/regulation/14628. Accessed July
16, 2013.
Dewatripont, M. and Legros, P. (2013). Essential patents, FRAND royalties and technological standards. The Journal of Industrial Economics, 61(4):913–937.
Dewenter, K. L. and Malatesta, P. H. (2001). State-owned and privately owned firms: An
empirical analysis of profitability, leverage, and labor intensity. American Economic
Review, 91(1):320–334.
Dewenter, R. and Haucap, J. (2006). Incentives to licence virtual mobile network operators (MVNOs). In Dewenter, R. and Haucap, J., editors, Access Pricing: Theory and
Practice. Elsevier, Amsterdam, Netherlands.
Dey, A. and Weis, S. (2010). Pseudoid: Enhancing privacy for federated login. In Hot
Topics in Privacy Enhancing Technologies (HotPETs) Proceedings, pages 95–107. Berlin,
Germany.
Diesing, P. (2008). Patterns of Discovery in the Social Sciences. AldineTransaction, New
Brunswick, NJ.
Dolbear, F. T., Lave, L. B., Bowman, G., Lieberman, A., Prescott, E., Rueter, F., and
Sherman, R. (1968). Collusion in oligopoly: An experiment on the effect of numbers
and information. The Quarterly Journal of Economics, 82(2):240–259.
Dorsey, R. E. (1992). The voluntary contributions mechanism with real time revisions.
Public Choice, 73(3):261–282.
Dosi, G. (1988). Sources, procedures, and microeconomic effects of innovation. Journal
of Economic Literature, 26(3):1120–71.
Dou, W. (2004). Will internet users pay for online content? Journal of Advertising Research, 44(4):349–359.
304
References
Drèze, X. and Hussherr, F.-X. (2003). Internet advertising: Is anybody watching? Journal
of Interactive Marketing, 17(4):8–23.
Duffy, J. (2006). Agent-based models and human subject experiments. In Tesfatsion, L.
and Judd, K., editors, Handbook of Computational Economics: Agent-based Computational
Economics, volume 2, pages 949–1011. North-Holland, Amsterdam, Netherlands.
Dufwenberg, M. and Gneezy, U. (2000). Price competition and market concentration:
an experimental study. International Journal of Industrial Organization, 18(1):7–22.
Duso, T., Neven, D. J., and Röller, L.-H. (2007). The political economy of European
merger control: Evidence using stock market data. The Journal of Law and Economics,
50(3):455–489.
Easley, R. F., Guo, H., and Krämer, J. (2015). From network neutrality to data neutrality:
A techno-economic framework and research agenda. Working Paper. Available at
http://ssrn.com/abstract=2666217.
Easterbrook, P. J., Gopalan, R., Berlin, J., and Matthews, D. R. (1991). Publication bias
in clinical research. The Lancet, 337(8746):867–872.
Economides, N. (1998). The incentive for non-price discrimination by an input monopolist. International Journal of Industrial Organization, 16(3):271–284.
Egelman, S. (2013). My profile is my password, verify me!: The privacy/convenience
tradeoff of Facebook Connect. In Proceedings of the SIGCHI Conference on Human Factors in Computing Systems, pages 2369–2378. Paris, France.
Einav, L. and Levin, J. (2010). Empirical industrial organization: A progress report.
Journal of Economic Perspectives, 24(2):145–162.
Eisenmann, T. R., Parker, G., and Van Alstyne, M. W. (2008). Opening platforms: How,
when and why? In Gawer, A., editor, Platforms, Markets & Innovation, pages 131–162.
Edwar Elgar Publishing, Cheltenham, UK.
305
References
eMarketer (2015). Facebook and Twitter will take 33% share of US digital display market by 2017. Available at http://www.emarketer.com/Article/FacebookTwitter-Will-Take-33-Share-of-US-Digital-Display-Market-by2017/1012274. Accessed March 02, 2016.
Engel, C. (2007). How much collusion? A meta-analysis of oligopoly experiments.
Journal of Competition Law & Economics, 3(4):491–549.
Erev, I. and Roth, A. (1998). Predicting how people play games: Reinforcement learning
in experimental games with unique, mixed strategy equilibria. American Economic
Review, 88(4):848–881.
Ergas, H., Ralph, E. K., and Lanigan, E. (2010). Price squeezes and imputation tests on
next generation access networks. Communications & Strategies, 78(2):97–85.
Errington, T. M., Iorns, E., Gunn, W., Tan, F. E., Lomax, J., and Nosek, B. A. (2014). An
open investigation of the reproducibility of cancer biology research. eLife, 3:e04333.
ETNO (2013). Study calls for new wave of European telecoms market deregulation
to restore growth and investment.
Available at http://www.etno.be/home/
press-corner/etno-press-releases/2013/216. Accessed July 16, 2013.
European Commission (2002). Directive 2002/19/EC of the European Parliament and
of the Council of 7 March 2002 on access to, and interconnection of, electronic communications networks and associated facilities (Access Directive). Official Journal of
the European Union, L 108:7–20.
European Commission (2008). Telecommunications: Commission raises serious doubts
about proposed Spanish broadband regulation. Available at http://europa.eu/
rapid/press-release_IP-08-1704_en.htm. Accessed May 05, 2016.
European Commission (2009a). Communication from the Commission - Guidance on
the Commission’s enforcement priorities in applying Article 82 of the EC Treaty to
306
References
abusive exclusionary conduct by dominant undertakings (2009/C 45/02). Official
Journal of the European Union, C 45:7–20.
European Commission (2009b). Community Guidelines for the application of state aid
rules in relation to rapid deployment of broadband networks (2009/C 235/04). Official Journal of the European Union, C 235:22–33.
European Commission (2010).
Communication from the Commission - Europe
2020 - A strategy for smart, sustainable and inclusive growth.
Available at
http://eur-lex.europa.eu/LexUriServ/LexUriServ.do?uri=COM:
2010:2020:FIN:EN:PDF. Accessed May 05, 2016.
European Commission (2011).
Guide to broadband investment.
Available at
http://ec.europa.eu/regional_policy/sources/docgener/presenta/
broadband2011/broadband2011_en.pdf. Accessed July 16, 2013.
European Commission (2012).
Mergers: Commission clears acquisition of Aus-
trian mobile phone operator Orange by H3G, subject to conditions. Available at
http://europa.eu/rapid/press-release_IP-12-1361_en.htm. Accessed
January 29, 2016.
European Commission (2013a).
Commission recommendation on consistent non-
discrimination obligations and costing methodologies to promote competition and
enhance the broadband investment environment (2013/466/EU). Official Journal of
the European Union, L 251:13–32.
European Commission (2013b). Commission staff working document. Digital agenda
scoreboard 2013. SWD(2013) 217. Available at http://ec.europa.eu/digitalagenda/en/download-scoreboard-reports. Accessed July 16, 2013.
European Commission (2013c). EU Guidelines for the application of state aid rules in
relation to the rapid deployment of broadband networks (2013/C 25/01). Official
Journal of the European Union, C 25:1–26.
307
References
European Commission (2013d). European Commission formally requests the Austrian regulator to amend or withdraw its proposal for broadband wholesale access
regulation. Available at http://europa.eu/rapid/press-release_IP-131172_en.htm. Accessed May 05, 2016.
European Commission (2013e). Proposal for a Regulation of the European Parliament and of the Council laying down measures concerning the European single market for electronic communications and to achieve a connected continent
- COM(2013) 627.
Available at https://ec.europa.eu/digital-single-
market/en/node/67489/. Accessed May 05, 2016.
European Commission (2013f). Telecoms - Commission suspends Austrian regulator’s proposal on broadband wholesale pricing. Available at http://europa.eu/
rapid/press-release_IP-13-748_en.htm. Accessed May 05, 2016.
European Commission (2014a). Commission recommendation on relevant product
and service markets within the electronic communications sector susceptible to ex
ante regulation in accordance with directive 2002/21/EC.
Available at https:
//ec.europa.eu/digital-single-market/en/news/commissionrecommendation-relevant-product-and-service-markets-withinelectronic-communications. Accessed May 05, 2016.
European Commission (2014b). Commission staff working document. Implementation
of the EU regulatory framework for electronic communications - 2014. SWD(2014)
249.
Available at http://ec.europa.eu/digital-agenda/en/download-
scoreboard-reports. Accessed December 10, 2015.
European Commission (2014c). Mergers: Commission clears acquisition of E-Plus by
Telefónica Deutschland, subject to conditions. Available at http://europa.eu/
rapid/press-release_IP-14-771_en.htm. Accessed January 29, 2016.
European Commission (2014d). Mergers: Commission clears acquisition of Telefónica
Ireland by Hutchison 3G, subject to conditions. Available at http://europa.eu/
308
References
rapid/press-release_IP-14-607_en.htm. Accessed January 29, 2016.
European Commission (2015a).
A Digital Single Market Strategy for Europe -
COM(2015) 192. Available at http://eur-lex.europa.eu/legal-content/
EN/TXT/?qid=1447773803386&uri=CELEX:52015DC0192. Accessed May 05,
2016.
European Commission (2015b). Antitrust: Commission sends Statement of Objections to Google on comparison shopping service; opens separate formal investigation
on Android. Available at http://europa.eu/rapid/press-release_IP-154780_en.htm. Accessed May 05, 2016.
European Commission (2015c).
Commission staff working document. A dig-
ital single market strategy for Europe - Analysis and evidence.
able
at
Avail-
http://eur-lex.europa.eu/legal-content/EN/TXT/?qid=
1447773803386&uri=CELEX:52015SC0100. Accessed May 05, 2016.
European Commission (2015d). Regulation (EU) 2015/2120 of the European Parliament
and of the Council of 25 November 2015 laying down measures concerning open
internet access and amending Directive 2002/22/EC on universal service and users’
rights relating to electronic communications networks and services and Regulation
(EU) No 531/2012 on roaming on public mobile communications networks within
the Union. Official Journal of the European Union, L 310:1–18.
European Commission (2016a). Antitrust: Commission sends Statement of Objections
to Google on Android operating system and applications. Available at http://
europa.eu/rapid/press-release_IP-16-1492_en.htm. Accessed May 05,
2016.
European Commission (2016b).
Background and definitions for the public con-
sultation on the needs for Internet speed and quality 2020.
Available at
https://ec.europa.eu/digital-single-market/en/news/public-
309
References
consultation-needs-internet-speed-and-quality-beyond-2020.
Accessed May 05, 2016.
European Commission (2016c). Synopsis report on the public consultation on the evaluation and review of the regulatory framework for electronic communications. Available at https://ec.europa.eu/digital-single-market/en/news/fullsynopsis-report-public-consultation-evaluation-and-reviewregulatory-framework-electronic. Accessed May 05, 2016.
Evans, D. (2008). The economics of the online advertising industry. Review of Network
Economics, 7(3):1–33.
Evans, D. (2009). The online advertising industry: Economics, evolution, and privacy.
The Journal of Economic Perspectives, 23(3):37–60.
Evans, D. and Schmalensee, R. (2009). The industrial organization of markets with
two-sided platforms. Competition Policy International, 3(1):151–179.
Evans, D. and Schmalensee, R. (2013). The antitrust analysis of multi-sided platform
businesses. In Blair, R. and Sokol, D., editors, Oxford Handbook on International Antitrust Economics, pages 404–450. Oxford University Press, Oxford, UK.
Ewing, B. T. and Kruse, J. B. (2010). An experimental examination of market concentration and capacity effects on price competition. Journal of Business Valuation and
Economic Loss Analysis, 5(1):1–14.
Facebook (2016). Company info - Stats. Available at http://newsroom.fb.com/
company-info/. Accessed March 02, 2016.
Falk, A. and Heckman, J. J. (2009). Lab experiments are a major source of knowledge in
the social sciences. Science, 326(5952):535–538.
Farrell, J. and Katz, M. L. (2000). Innovation, rent extraction, and integration in systems
markets. The Journal of Industrial Economics, 48(4):413–432.
310
References
Farrell, J. and Weiser, P. J. (2003). Modularity, vertical integration, and open access policies: Towards a convergence of antitrust and regulation in the internt age. Harvard
Journal of Law & Technology, 17(1):85–134.
Federal Communications Commission (2011). WT Docket No. 11-65 – Staff analysis and
findings. Available at https://apps.fcc.gov/edocs_public/attachmatch/
DA-11-1955A2.pdf. Accessed January 29, 2016.
Federal Communications Commission (2014).
Statement from FCC chair-
man Tom Wheeler on competition in the mobile marketplace.
Avail-
able at http://www.fcc.gov/document/chairman-wheeler-statementcompetition-mobile-marketplace. Accessed January 29, 2016.
Feeley, T. H., Tutzauer, F., Rosenfeld, H. L., and Young, M. J. (1997). Cooperation in an
infinite-choice continuous-time prisoner’s dilemma. Simulation & Gaming, 28(4):442–
459.
Fehr, E. and Schmidt, K. M. (1999). A theory of fairness, competition, and cooperation.
The Quarterly Journal of Economics, 114(3):817–868.
Feingold, S. (2013). The hidden risks behind Facebook’s login tools. Available at
http://www.ktsrecruits.com/en/Knowledge_Center/Publications/
Articles/2013/01/The_Hidden_Risks_Behind_Facebooks_Login_
Tools.aspx. Accessed March 02, 2016.
Fonseca, M. A., Huck, S., and Normann, H.-T. (2005). Playing Cournot although they
shouldn’t. Economic Theory, 25(3):669–677.
Fonseca, M. A. and Normann, H.-T. (2008). Mergers, asymmetries and collusion: Experimental evidence. The Economic Journal, 118(527):387–400.
Fonseca, M. A. and Normann, H.-T. (2012). Explicit vs. tacit collusion - The impact
of communication in oligopoly experiments. European Economic Review, 56(8):1759–
311
References
1772.
Ford, G. S. (2007). Does a municipal electric’s supply of communications crowd out private communications investment? An empirical study. Energy Economics, 29(3):467–
478.
Forzati, M., Larsen, C. P., and Mattsson, C. (2010). Open access networks, the Swedish
experience. In 12th International Conference on Transparent Optical Networks (ICTON
2010), pages 1–4. Munich, Germany.
Fouraker, L. E. and Siegel, S. (1963). Bargaining Behavior. McGraw-Hill, New York, NY.
Fournier, S. and Avery, J. (2011). The uninvited brand. Business Horizons, 54(3):193–207.
Frank, U. (2006). Towards a pluralistic conception of research methods in Information Systems Research. Working Paper. Available at http://www.econstor.eu/
obitstream/10419/58156/1/715932098.pdf.
Freeman, J. B. and Ambady, N. (2010). Mousetracker: Software for studying real-time
mental processing using a computer mouse-tracking method. Behavior Research Methods, 42(1):226–241.
Friedman, D., Huck, S., Oprea, R., and Weidenholzer, S. (2015). From imitation to collusion: Long-run learning in a low-information environment. Journal of Economic
Theory, 155:185–205.
Friedman, D. and Oprea, R. (2012). A continuous dilemma. American Economic Review,
102(1):337–363.
Friedman, J. W. (1971). A non-cooperative equilibrium for supergames. The Review of
Economic Studies, 38(1):1–12.
Friedman, M. (1953). Essays In Positive Economics. University Press, Chicago, IL.
312
References
Fudenberg, D., Rand, D. G., and Dreber, A. (2012). Slow to anger and fast to forgive:
Cooperation in an uncertain world. American Economic Review, 102(2):720–749.
Fudenberg, D. and Tirole, J. (1984). The fat-cat effect, the puppy-dog ploy, and the lean
and hungry look. American Economic Review, 74(2):361–66.
Fudenberg, D., Tirole, J., et al. (2000). Customer poaching and brand switching. The
RAND Journal of Economics, 31(4):634–657.
Fudenberg, D. and Villas-Boas, J. M. (2007). Behavior-based price discrimination and
customer recognition. In Hendershott, T., editor, Handbook on Economics and Information Systems, pages 377–435. Elsevier Science, Oxford, UK.
Fudenberg, D. and Villas-Boas, J. M. (2012). Price discrimination in the digital economy.
In Peitz, M. and Waldfogel, J., editors, The Oxford Handbook of the Digital Economy.
Oxford University Press, New York, NY.
Gafni, R. and Nissim, D. (2014). To social login or not login? Exploring factors affecting
the decision. Issues in Informing Science and Information Technology, 11:57–72.
Gans, J. and King, S. (2003). Access holidays for network infrastructure investment.
Agenda, 10(2):163–178.
Gans, J. S. (2007). Vertical contracting when competition for orders precedes procurement. The Journal of Industrial Economics, 55(2):325–346.
Gans, J. S. and King, S. P. (2004). Access holidays and the timing of infrastructure
investment. Economic Record, 80(248):89–100.
Garcia, S., Moreaux, M., and Reynaud, A. (2007). Measuring economies of vertical
integration in network industries: An application to the water sector. International
Journal of Industrial Organization, 25(4):791–820.
313
References
Garella, P. G. and Petrakis, E. (2008). Minimum quality standards and consumers’ information. Economic Theory, 36(2):283–302.
Gaudin, G. and Mantzari, D. (2016). Margin squeeze: An above-cost predatory pricing
approach. Journal of Competition Law & Economics, 12(1):151–179.
Gaudin, G. and Saavedra, C. (2014). Ex ante margin squeeze tests in the telecommunications industry: What is a reasonably efficient operator? Telecommunications Policy,
38(2):157–172.
Gayle, P. G. and Weisman, D. L. (2007). Are input prices irrelevant for make-or-buy
decisions? Journal of Regulatory Economics, 32(2):195–207.
Genakos, C., Valletti, T., and Verboven, F. (2015). Evaluating market consolidation in
mobile communications. Available at http://cerre.eu/sites/cerre/files/
150915_CERRE_Mobile_Consolidation_Report_Final.pdf. Accessed May
05, 2016.
Geradin, D. and O’Donoghue, R. (2005). The concurrent application of competition law
and regulation: The case of margin squeeze abuses in the telecommunications sector.
Journal of Competition Law & Economics, 1(2):355–425.
Geroski, P. A. (2003). Competition in markets and competition for markets. Journal of
Industry, Competition and Trade, 3(3):151–166.
Gibbard, A. and Varian, H. R. (1978). Economic models. The Journal of Philosophy,
75(11):664–677.
Gigya (2015). Social login 101: Everything you need to know about social login and
the future of customer identity. Available at http://www.gigya.com/resource/
whitepaper/social-login-101/. Accessed March 02, 2016.
Gilbert, R. J. (2006). Competition and innovation. Journal of Industrial Organization
Education, 1(1):1–23.
314
References
Gilboa, I., Postlewaite, A., Samuelson, L., and Schmeidler, D. (2014). Economic models
as analogies. The Economic Journal, 124(578):513–533.
Gillett, S. E., Lehr, W. H., and Osorio, C. A. (2006). Municipal electric utilities’ role in
telecommunications services. Telecommunications Policy, 30(8):464–480.
Gimpel, H., Jennings, N. R., Kersten, G. E., Ockenfels, A., and Weinhardt, C. (2008).
Market engineering: A research agenda. In Negotiation, Auctions, and Market Engineering. Volume 2 of Lecture Notes in Business Information Processing, pages 1–15. Springer,
Berlin, Germany.
Given, J. (2010). Take your partners: Public private interplay in Australian and New
Zealand plans for next generation broadband. Telecommunications Policy, 34(9):540–
549.
Goldfarb, A. and Tucker, C. (2011a). Online display advertising: Targeting and obtrusiveness. Marketing Science, 30(3):389–404.
Goldfarb, A. and Tucker, C. E. (2011b). Privacy regulation and online advertising. Management Science, 57(1):57–71.
Gómez-Barroso, J. L. and Feijóo, C. (2010). A conceptual framework for public-private
interplay in the telecommunications sector. Telecommunications Policy, 34(9):487–495.
Gonçalves, R. (2007). Cost orientation and xDSL services: Retail-minus vs. LRAIC.
Telecommunications Policy, 31(8):524–529.
Gonçalves, R. and Nascimento, Á. (2010). The momentum for network separation: A
guide for regulators. Telecommunications Policy, 34(7):355–365.
Goren, H., Kurzban, R., and Rapoport, A. (2003). Social loafing vs. social enhancement:
Public goods provisioning in real-time with irrevocable commitments. Organizational
Behavior and Human Decision Processes, 90(2):277–290.
315
References
Goren, H., Rapoport, A., and Kurzban, R. (2004). Revocable commitments to public
goods provision under the real-time protocol of play. Journal of Behavioral Decision
Making, 17(1):17–37.
Götte, L. and Schmutzler, A. (2009). Merger policy: What can we learn from experiments? In Hinloopen, J. and Normann, H.-T., editors, Experiments and Competition
Policy, pages 185–216. Cambridge University Press, Cambridge, UK.
Graef, I., Wahyuningtyas, S. Y., and Valcke, P. (2015). Assessing data access issues in
online platforms. Telecommunications Policy, 39(5):375–387.
Gregor, S. (2006).
The nature of theory in Information Systems.
MIS Quarterly,
30(3):611–642.
Greiner, B. (2015). Subject pool recruitment procedures: Organizing experiments with
ORSEE. Journal of the Economic Science Association, 1(1):114–125.
Griesemer, J. (2013). Formalization and the meaning of “theory” in the inexact biological sciences. Biological Theory, 7(4):298–310.
Grove, J. V. (2010). Each month 250 million people use Facebook Connect on the
web. Available at http://mashable.com/2010/12/08/facebook-connectstats/. Accessed on 02.03.2016.
Grzybowski, L. and Liang, J. (2014). Estimating demand for quadruple-play tariffs: The
impact on consumer surplus. In 25th European Regional Conference of the International
Telecommunications Society (ITS). Brussels, Belgium.
Guala, F. (2005). The Methodology of Experimental Economics. Cambridge University
Press, New York, NY.
Güth, W., Levati, M. V., and Stiehler, A. (2002). Privately contributing to public goods
over time: An experimental study.
handle.net/10419/65343.
316
Working Paper. Available at http://hdl.
References
Guthrie, G. (2006). Regulating infrastructure: The impact on risk and investment. Journal of Economic Literature, 44(4):925–972.
Häckner, J. (2000).
A note on price and quantity competition in differentiated
oligopolies. Journal of Economic Theory, 93(2):233–239.
Hagiu, A. and Wright, J. (2015). Multi-sided platforms. International Journal of Industrial
Organization, 43:162–174.
Hambrick, D. C. (2007). The field of management’s devotion to theory: Too much of a
good thing? Academy of Management Journal, 50(6):1346–1352.
Hamman, D. and Plomion, B. (2013). Chango retargeting barometer. Available at
http://www.iab.net/media/file/Chango_Retargeting_Barometer_
April_2013.pdf. Accessed March 02, 2016.
Harbord, R. M. and Higgins, J. P. T. (2008). Meta-regression in Stata. The Stata Journal,
8(4):493–519.
Hardt, M. (1995). The non-equivalence of accounting separation and structural separation as regulatory devices. Telecommunications Policy, 19(1):69–72.
Harré, R. (2003). The materiality of instruments in a metaphysics for experiments. In
Radder, H., editor, The Philosophy Of Scientific Experimentation, pages 19–38. University of Pittsburgh Press, Pittsburgh, PA.
Hart, O. (2003). Incomplete contracts and public ownership: Remarks, and an application to public-private partnerships. The Economic Journal, 113(486):69–76.
Hart, O. and Tirole, J. (1990). Vertical integration and market foreclosure. Brookings
Papers on Economic Activity: Microeconomics, pages 205–286.
Hashmi, A. R. (2013). Competition and innovation: The inverted-U relationship revisited. The Review of Economics and Statistics, 95(5):1653–1668.
317
References
Hastie, T., Tibshirani, R., Friedman, J., and Franklin, J. (2009). The Elements of Statistical
Learning: Data Mining, Inference, and Prediction. Springer, New York, NY.
Haucap, J. and Heimeshoff, U. (2014). Google, Facebook, Amazon, eBay: Is the internet
driving competition or market monopolization? International Economics and Economic
Policy, 11(1):49–61.
Hauge, J. A., Jamison, M. A., and Gentry, R. J. (2008). Bureaucrats as entrepreneurs: Do
municipal telecommunications providers hinder private entrepreneurs? Information
Economics and Policy, 20(1):89–102.
Hausman, D. M. (1990). Supply and demand explanations and their ceteris paribus
clauses. Review of Political Economy, 2(2):168–187.
Hausman, D. M. (2013). Philosophy of Economics. In Zalta, E. N., editor, The Stanford
Encyclopedia of Philosophy. Center for the Study of Language and Information, Stanford University, Stanford, CA, Winter 2013 edition.
Hawkins, R. X. D. (2015). Conducting real-time multiplayer experiments on the web.
Behavior Research Methods, 47(4):966–976.
Heimler, A. (2010). Is a margin squeeze an antitrust or a regulatory violation? Journal
of Competition Law & Economics, 6(4):879–890.
Hellwig, M. F. (2008). Competition policy and sector-specific regulation for network
industries. Working Paper. Available at http://ssrn.com/abstract=1275285.
Hevner, A. R., March, S. T., Park, J., and Ram, S. (2004). Design science in Information
Systems Research. MIS Quarterly, 28(1):75–105.
Higgins, J. and Thompson, S. G. (2002). Quantifying heterogeneity in a meta-analysis.
Statistics in Medicine, 21(11):1539–1558.
318
References
Hodge, G. A. and Greve, C. (2007). Public–private partnerships: An international performance review. Public Administration Review, 67(3):545–558.
Hoernig, S., Jay, S., Neumann, K.-H., Peitz, M., Plückebaum, T., and Vogelsang, I. (2012).
The impact of different fibre access network technologies on cost, competition and
welfare. Telecommunications Policy, 36(2):96–112.
Höffler, F. (2007). Cost and benefits from infrastructure competition. Estimating welfare
effects from broadband access competition. Telecommunications Policy, 31(6):401–418.
Höffler, F. (2008). On the consistent use of linear demand systems if not all varieties are
available. Economics Bulletin, 4(14):1–5.
Höffler, F. and Schmidt, K. M. (2008). Two tales on resale. International Journal of Industrial Organization, 26(6):1448–1460.
Hogendorn, C. (2007). Broadband internet: net neutrality versus open access. International Economics and Economic Policy, 4(2):185–208.
Horstmann, N., Krämer, J., and Schnurr, D. (2015). Oligopoly competition in continuous
time. Working Paper. Available at http://ssrn.com/abstract=2630664.
Horstmann, N., Krämer, J., and Schnurr, D. (2016a).
Margin squeeze regulation
and infrastructure competition. Working Paper. Available at http://ssrn.com/
abstract=2791575.
Horstmann, N., Krämer, J., and Schnurr, D. (2016b). Number effects in oligopolies: How
many competitors are enough to ensure competition? Working Paper. Available at
http://ssrn.com/abstract=2535862.
Horstmann, N., Krämer, J., and Schnurr, D. (2016c). Wholesale competition and open
access regimes: Experimental evidence. Working Paper. Available at http://ssrn.
com/abstract=2630660.
319
References
Hossain, T., Minor, D., and Morgan, J. (2011). Competing matchmakers: An experimental analysis. Management Science, 57(11):1913–1925.
Hotelling, H. (1929). Stability in competition. The Economic Journal, 39(153):41–57.
Houy, C., Fettke, P., and Loos, P. (2015). Stylized facts as an instrument for literature
review and cumulative Information Systems Research. Communications of the Association for Information Systems, 37(1):225–256.
Hu, Y., Shin, J., and Tang, Z. (2015). Incentive problems in performance-based online
advertising pricing: Cost per click vs. cost per action. Management Science. Article in
advance. doi:10.1287/mnsc.2015.2223.
Huck, S., Müller, W., and Normann, H.-T. (2001). Stackelberg beats cournot: On collusion and efficiency in experimental markets. The Economic Journal, pages 749–765.
Huck, S., Normann, H.-T., and Oechssler, J. (1999). Learning in Cournot oligopoly - An
experiment. The Economic Journal, 109(454):80–95.
Huck, S., Normann, H.-T., and Oechssler, J. (2004a). Through trial and error to collusion.
International Economic Review, 45(1):205–224.
Huck, S., Normann, H.-T., and Oechssler, J. (2004b). Two are few and four are many:
number effects in experimental oligopolies. Journal of Economic Behavior & Organization, 53(4):435–446.
Igami, M. and Uetake, K. (2015). Mergers, innovation, and entry-exit dynamics: The
consolidation of the hard disk drive industry (1976-2014). Working Paper. Available
at http://ssrn.com/abstract=2585840.
Inderst, R. and Peitz, M. (2012). Network investment, access and competition. Telecommunications Policy, 36(5):407–418.
320
References
Inderst, R. and Valletti, T. (2009). Price discrimination in input markets. The RAND
Journal of Economics, 40(1):1–19.
Ioannidis, J. P. A. (2005). Why most published research findings are false. PLoS Med,
2(8):696–701.
Iossa, E. and Stroffolini, F. (2012). Vertical integration and costly demand information
in regulated network industries. Review of Industrial Organization, 40(4):249–271.
Ishii, K. and Kurzban, R. (2008). Public goods games in Japan. Human Nature, 19(2):138–
156.
IT-Summit (2010). Ergebnispapier der Projektgruppe “Open Access” der AG 2 im Nationalen IT-Gipfel. Available at http://www.bmwi.de/Dateien/BMWi/PDF/ITGipfel/it-gipfel-2010-open-access,property=pdf,bereich=
bmwi2012,sprache=en,rwb=true.pdf. Accessed July 16, 2013.
Ivaldi, M., Jullien, B., Rey, P., Seabright, P., and Tirole, J. (2003).
The economics
of tacit collusion. Report for DG Competition, European Commission. Available
at http://ec.europa.eu/competition/mergers/studies_reports/the_
economics_of_tacit_collusion_en.pdf. Accessed May 05, 2016.
Iyer, G., Soberman, D., and Villas-Boas, J. M. (2005). The targeting of advertising. Marketing Science, 24(3):461–476.
Janrain (2014). Social login and personalization - 2014 Janrain US consumer research.
Available at http://janrain.com/resources/industry-research/2014consumer-research-social-login-personalization/. Accessed March
02, 2016.
Janssen, M. C. and Mendys-Kamphorst, E. (2008). Triple play: How do we secure future
benefits? Telecommunications Policy, 32(11):735–743.
321
References
Jara-Dıaz, S., Ramos-Real, F. J., and Martınez-Budrıa, E. (2004). Economies of integration in the Spanish electricity industry using a multistage cost function. Energy
Economics, 26(6):995–1013.
Jaspers, F. and van den Ende, J. (2006). The organizational form of vertical relationships:
Dimensions of integration. Industrial Marketing Management, 35(7):819–828.
Jaunaux, L. and Lebourges, M. (2015). Economic replicability tests for next-generation
access networks. Telecommunications Policy, 39(6):486–501.
Jiang, Z. and Sarkar, S. (2009). Speed matters: The role of free software offer in software
diffusion. Journal of Management Information Systems, 26(3):207–240.
Johnson, J. P. (2013). Targeted advertising and advertising avoidance. The RAND Journal
of Economics, 44(1):128–144.
Jordan, S. (2009). Implications of internet architecture on net neutrality. ACM Transactions on Internet Technology, 9(2):1–28.
Jorde, T. M. and Teece, D. J. (1990). Innovation and cooperation: Implications for competition and antitrust. Journal of Economic Perspectives, 4(3):75–96.
Jorgenson, D. W. and Vu, K. M. (2016). The ICT revolution, world economic growth,
and policy issues. Telecommunications Policy, 40(5):383–397.
Jullien, B., Rey, P., and Saavedra, C. (2014). The economics of margin squeeze. Working Paper. Available at http://idei.fr/doc/by/jullien/Margin_Squeeze_
Policy_Paper_revised_March_2014.pdf.
Kalmus, P. and Wiethaus, L. (2010). On the competitive effects of mobile virtual network operators. Telecommunications Policy, 34(5):262–269.
Kaserman, D. L. and Mayo, J. W. (1991). The measurement of vertical economies and
the efficient structure of the electric utility industry. The Journal of Industrial Economics,
322
References
39(5):483–502.
Kazovsky, L. G., Shaw, W.-T., Gutierrez, D., Cheng, N., and Wong, S.-W. (2007). Nextgeneration optical access networks. Journal of Lightwave Technology, 25(11):3428–3442.
Kephart, C. and Friedman, D. (2015). Hotelling revisits the lab: equilibration in continuous and discrete time. Journal of the Economic Science Association, 1(2):132–145.
Kephart, C. and Rose, L. (2015). Stability in competition? Hotelling in continuous
time. Working Paper. Available at http://www.cazaar.com/home/research/
KephartRose_Hotelling_PL.pdf.
Kerlinger, F. N. (1986). Foundations of Behavioral Research. Holt, Rinehart and Winston,
Fort Worth, TX.
Ketcham, J., Smith, V. L., and Williams, A. W. (1984). A comparison of posted-offer and
double-auction pricing institutions. The Review of Economic Studies, 51(4):595–614.
Ketter, W., Collins, J., and Reddy, P. (2013). Power TAC: A competitive economic simulation of the smart grid. Energy Economics, 39:262–270.
Kitchin, R. (2014). The Data Revolution - Big Data, Open Data, Data Infrastructures and
Their Consequences. SAGE, London, UK.
Klein, R. A., Ratliff, K. A., Vianello, M., Jr., R. B. A., Bahnik, S., Bernstein, M. J., Bocian,
K., Brandt, M. J., et al. (2014). Investigating variation in replicability: A “many labs”
replication project. Social Psychology, 45(3):142–152.
Klumpp, T. and Su, X. (2010). Open access and dynamic efficiency. American Economic
Journal: Microeconomics, 2(2):64–96.
Klumpp, T. and Su, X. (2015). Strategic investment under open access: Theory and
evidence. The Journal of Industrial Economics, 63(3):495–521.
323
References
Knigge, A. and Buskens, V. (2010). Coordination and cooperation problems in network
good production. Games, 1(4):357–380.
Ko, M. N., Cheek, G. P., Shehab, M., and Sandhu, R. (2010). Social-networks connect
services. Computer, 43(8):37–43.
Koenker, R. and Hallock, K. (2001). Quantile regression: An introduction. The Journal of
Economic Perspectives, 15(4):43–56.
Kolasky, W. J. and Dick, A. R. (2003). The merger of guidelines and the integration of efficiencies into antitrust review of horizontal mergers. Antitrust Law Journal, 71(1):207–
251.
Kondaurova, I. and Weisman, D. L. (2003). Incentives for non-price discrimination.
Information Economics and Policy, 15(2):147–171.
Kontaxis, G., Polychronakis, M., and Markatos, E. P. (2012). Minimizing information
disclosure to third parties in social login platforms. International Journal of Information
Security, 11(5):321–332.
Koopmans, T. C. (1957). Three essays on the state of economic analysis. McGraw-Hill, New
York, NY.
Kourandi, F., Krämer, J., and Valletti, T. (2015). Net neutrality, exclusivity contracts, and
internet fragmentation. Information Systems Research, 26(2):320–338.
Krämer, J. and Schnurr, D. (2014). A unified framework for open access regulation
of telecommunications infrastructure: Review of the economic literature and policy
guidelines. Telecommunications Policy, 38(11):1160–1179.
Krämer, J. and Schnurr, D. (2016). Microeconomically founded Information Systems
Research. Business & Information Systems Engineering, 48(4). Article in advance.
324
References
Krämer, J., Schnurr, D., and Wohlfarth, M. (2016). Winners, losers, and Facebook: The
role of social logins in the online advertising ecosystem. Working Paper. Available at
http://ssrn.com/abstract=2791591.
Krämer, J. and Vogelsang, I. (2016). Co-investments and tacit collusion in regulated
network industries: Experimental evidence. Working Paper. Available at http://
ssrn.com/abstract=2119927.
Krämer, J., Wiewiorra, L., and Weinhardt, C. (2013). Net neutrality: A progress report.
Telecommunications Policy, 37(9):794–813.
Krämer, J. and Wohlfarth, M. (2015). Regulating over-the-top service providers in twosided content markets: Insights from the economic literature. Communications &
Strategies, 99(3):71–90.
Kroes, N. (2012). Enhancing the broadband investment environment. Available at
http://europa.eu/rapid/press-release_SPEECH-12-552_en.htm.
Ac-
cessed July 16, 2013.
Kübler, D. and Müller, W. (2002). Simultaneous and sequential price competition in
heterogeneous duopoly markets: Experimental evidence. International Journal of Industrial Organization, 20(10):1437–1460.
Kurzban, R., McCabe, K., Smith, V. L., and Wilson, B. J. (2001). Incremental commitment
and reciprocity in a real-time public goods game. Personality and Social Psychology
Bulletin, 27(12):1662–1673.
Kwoka, J. E. (2002). Vertical economies in electric power: Evidence on integration and
its alternatives. International Journal of Industrial Organization, 20(5):653–671.
Lachin, J. M. (1981). Introduction to sample size determination and power analysis for
clinical trials. Controlled Clinical Trials, 2(2):93–113.
325
References
Laffont, J. J. and Tirole, J. (2001). Competition in Telecommunications. MIT Press, Cambridge, MA.
Lafontaine, F. and Slade, M. (2007). Vertical integration and firm boundaries: The evidence. Journal of Economic Literature, 45(3):629–685.
Lambrecht, A. and Tucker, C. (2013). When does retargeting work? Information specificity in online advertising. Journal of Marketing Research, 50(5):561–576.
Lawson, T. (2008). What has realism got to do with it? In Hausman, D. M., editor,
The Philosophy of Economics: An Anthology, pages 439–453. Cambridge Univ Press,
Cambridge, UK, third edition.
Lehr, W., Sirbu, M., and Gillett, S. (2004). Broadband open access: Lessons from municipal network case studies. In Telecommunications Policy Research Conference (TPRC).
Arlington, VA.
Lenth, R. V. (2001). Some practical guidelines for effective sample size determination.
The American Statistician, 55(3):187–193.
Levati, M. V. and Neugebauer, T. (2004). An application of the English clock market
mechanism to public goods games. Experimental Economics, 7(2):153–169.
Levin, J. D. (2013). The economics of internet markets. In Acemoglu, D., Arellano, M.,
and Dekel, E., editors, Advances in Economics and Econometrics. Cambridge University
Press, New York, NY.
Lewis, R. A. and Reiley, D. H. (2014). Online ads and offline sales: Measuring the effect
of retail advertising via a controlled experiment on yahoo! Quantitative Marketing and
Economics, 12(3):235–266.
Leymann, N., Heidemann, C., Wasserman, M., Xue, L., and Zhang, M. (2015). Hybrid
access network architecture. Available at https://tools.ietf.org/id/draft-
326
References
lhwxz-hybrid-access-network-architecture-02.txt. Accessed May 05,
2016.
Linder, S. H. and Rosenau, P. V. (2000). Mapping the terrain of the public-private policy
partnership. In Rosenau, P. V., editor, Public-private policy partnerships, pages 1–18.
MIT Press, Cambridge, MA.
Lindqvist, T. and Stennek, J. (2005). The insiders’ dilemma: An experiment on merger
formation. Experimental Economics, 8(3):267–284.
List, J. A., Sadoff, S., and Wagner, M. (2011). So you want to run an experiment, now
what? Some simple rules of thumb for optimal experimental design. Experimental
Economics, 14(4):439–457.
Liu, D. and Viswanathan, S. (2014). Information asymmetry and hybrid advertising.
Journal of Marketing Research, 51(5):609–624.
Locke, E. A. (2007). The case for inductive theory building. Journal of Management,
33(6):867–890.
LoginRadius (2015).
port.
Social login and social sharing trends - Q3 2015 Re-
Available at http://social.loginradius.com/social-login-
social-sharing-trends-q3-2015. Accessed March 02, 2016.
López, M. C. and Naylor, R. A. (2004). The Cournot–Bertrand profit differential: a
reversal result in a differentiated duopoly with wage bargaining. European Economic
Review, 48(3):681–696.
Lucas, R. E. (1980). Methods and problems in business cycle theory. Journal of Money,
Credit and Banking, 12(4):696–715.
Lunn, P. (2014). Regulatory Policy and Behavioural Economics. OECD Publishing, Paris,
France. doi:10.1787/9789264207851-en.
327
References
Mäki, U. (1992). On the method of isolation in Economics. In Dilworth, C., editor,
Idealization IV: Intelligbility in Science. Rodopi, Amsterdam, Netherlands.
Mäki, U. (2012). Realism and antirealism about Economics. In Mäki, U., editor, Philosophy of Economics, volume 13 of Handbook of the Philosophy of Science, pages 3–24.
Elsevier, Amsterdam, Netherlands.
Manchanda, P., Dubé, J.-P., Goh, K. Y., and Chintagunta, P. K. (2006). The effect of
banner advertising on internet purchasing. Journal of Marketing Research, 43(1):98–
108.
Mandy, D. and Sappington, D. (2007). Incentives for sabotage in vertically related industries. Journal of Regulatory Economics, 31(3):235–260.
Mandy, D. M. (2000). Killing the goose that may have laid the golden egg: Only the
data know whether sabotage pays. Journal of Regulatory Economics, 17(2):157–172.
Mankiw, N. G. (1986). Free entry and social inefficiency. The RAND Journal of Economics,
17(1):48–58.
Mantena, R. and Saha, R. L. (2012). Co-opetition between differentiated platforms in
two-sided markets. Journal of Management Information Systems, 29(2):109–140.
Mantovani, A. and Ruiz-Aliseda, F. (2015). Equilibrium innovation ecosystems: The
dark side of collaborating with complementors. Management Science, 62(2).
Marshall (2015). Facebook to boost mobile-ad market share, as eMarketer reverses
forecast.
The Wall Street Journal.
Available at http://blogs.wsj.com/cmo/
2015/09/08/facebook-projected-to-narrow-mobile-ad-gap-withgoogle-as-emarketer-reverses-forecast/. Accessed March 20, 2016.
Martin, S., Normann, H.-T., and Snyder, C. M. (2001). Vertical foreclosure in experimental markets. The RAND Journal of Economics, 32(3):466–96.
328
References
Maskin, E. and Tirole, J. (1988a). A theory of dynamic oligopoly, I: Overview and quantity competition with large fixed costs. Econometrica, 56(3):549–569.
Maskin, E. and Tirole, J. (1988b). A theory of dynamic oligopoly, II: Price competition,
kinked demand curves, and edgeworth cycles. Econometrica, 56(3):571–599.
Mason, C., Phillips, O. R., and Nowell, C. (1992). Duopoly behavior in asymmetric markets: An experimental evaluation. The Review of Economics and Statistics, 74(4):662–70.
McGowan, L. and Cini, M. (1999). Discretion and politicization in EU competition
policy: The case of merger control. Governance, 12(2):175–200.
McKelvey, R. D. and Palfrey, T. R. (1995). Quantal response equilibria for normal form
games. Games and Economic Behavior, 10(1):6–38.
Megginson, W. L. and Netter, J. M. (2001). From state to market: A survey of empirical
studies on privatization. Journal of Economic Literature, 39(2):321–389.
Midgley, D. F., Marks, R. E., and Cooper, L. C. (1997). Breeding competitive strategies.
Management Science, 43(3):257–275.
Millner, E. L., Pratt, M. D., and Reilly, R. J. (1990). Contestability in real-time experimental flow markets. The RAND Journal of Economics, 21(4):584–599.
Monopolkommission (2015).
kets.
Competition policy: The challenge of digital mar-
Available at http://www.monopolkommission.de/images/PDF/SG/
s68_fulltext_eng.pdf. Accessed May 05, 2016.
Montes, R., Sand-Zantman, W., and Valletti, T. M. (2015). The value of personal information in markets with endogenous privacy. Working Paper. Available at http:
//ssrn.com/abstract=2640177.
Morgan, M. S. (2002). Model experiments and models in experiments. In Magnani,
L. and Nersessian, N. J., editors, Model-Based Reasoning: Science, Technology, Values,
329
References
pages 41–58. Springer, New York, NY.
Morgan, M. S. (2005). Experiments versus models: New phenomena, inference and
surprise. Journal of Economic Methodology, 12(2):317–329.
Morgan, M. S. and Knuuttila, T. (2012). Models and modelling in Economics. In Mäki,
U., editor, Philosophy of Economics, volume 13 of Handbook of the Philosophy of Science,
pages 49–87. Elsevier, Amsterdam, Netherlands.
Morgan, M. S. and Morrison, M. (1999). Models as Mediators: Perspectives on Natural and
Social Science. Cambridge University Press, Cambridge, UK.
Müller, M. B., Hariharan, A., and Adam, M. T. P. (2014). A NeuroIS platform for lab
experiments. In Davis, F., Riedl, R., vom Brocke, J., Léger, P.-M., and Randolph, A.,
editors, Proceedings Gmunden Retreat on NeuroIS, pages 15–17. Gmunden, Austria.
Müller, W. (2006). Allowing for two production periods in the Cournot duopoly: Experimental evidence. Journal of Economic Behavior & Organization, 60(1):100–111.
Murphy, R. O., Ackermann, K. A., and Handgraaf, M. (2011). Measuring social value
orientation. Judgment and Decision Making, 6(8):771–781.
Murphy, R. O., Rapoport, A., and Parco, J. E. (2006). The breakdown of cooperation in
iterative real-time trust dilemmas. Experimental Economics, 9(2):147–166.
Nalebuff, B. J. and Brandenburger, A. M. (1997). Co-opetition: Competitive and cooperative business strategies for the digital economy. Strategy & Leadership, 25(6):28–33.
Nemoto, J. and Goto, M. (2004). Technological externalities and economies of vertical
integration in the electric utility industry. International Journal of Industrial Organization, 22(1):67–81.
Newman, N. (2014). Search, antitrust, and the economics of the control of user data.
Yale Journal on Regulation, 31(3):401–454.
330
References
Nicas, J. (2016).
nal.
FTC extends probe into Google’s Android.
Wall Street Jour-
Available at http://www.wsj.com/articles/ftc-extends-probe-
into-googles-android-1461699217. Accessed on 05.05.2016.
Nitsche, R. and Wiethaus, L. (2011). Access regulation and investment in next generation networks - A ranking of regulatory regimes. International Journal of Industrial
Organization, 29(2):263–272.
Noam, E. M. (2010). Regulation 3.0 for telecom 3.0. Telecommunications Policy, 34(1):4–10.
Nocke, V. and White, L. (2007). Do vertical mergers facilitate upstream collusion? American Economic Review, 97(4):1321–1339.
Normann, H.-T. (2009). Vertical integration, raising rivals’ costs and upstream collusion. European Economic Review, 53(4):461–480.
Normann, H. T. (2011).
Vertical mergers, foreclosure and raising rivals’ costs–
experimental evidence. The Journal of Industrial Economics, 59(3):506–527.
Normann, H.-T. and Ricciuti, R. (2009). Laboratory experiments for economic policy
making. Journal of Economic Surveys, 23(3):407–432.
North, D. C. (1991). Institutions. Journal of Economic Perspectives, 5(1):97–112.
Nosek, B. A., Alter, G., Banks, G. C., Borsboom, D., Bowman, S. D., Breckler, S. J.,
Buck, S., Chambers, C. D., et al. (2015). Promoting an open research culture. Science,
348(6242):1422–1425.
OECD (2013). Broadband networks and open access. OECD Digital Economy Papers,
No. 218. doi: 10.1787/5k49qgz7crmr-en.
OECD (2015). Triple and quadruple play bundles of communication services. OECD
Science, Technology and Industry Policy Papers, No. 23. doi: 10.1787/5js04dp2q1jc-en.
331
References
Oechssler, J., Roomets, A., and Roth, S. (2016). From imitation to collusion: a replication.
Journal of the Economic Science Association, 2(1):13–21.
Ofcom (2014).
Review of the wholesale broadband access markets.
Avail-
able at http://stakeholders.ofcom.org.uk/binaries/consultations/
review-wba-markets/statement/WBA-Statement.pdf. Accessed January
29, 2016.
Open Science Collaboration (2015). Estimating the reproducibility of psychological science. Science, 349(6251):1–8.
Oprea, R., Charness, G., and Friedman, D. (2014). Continuous time and communication
in a public-goods experiment. Journal of Economic Behavior & Organization, 108:212–
223.
Ordover, J. and Shaffer, G. (2007). Wholesale access in multi-firm markets: When is
it profitable to supply a competitor? International Journal of Industrial Organization,
25(5):1026–1045.
Ordover, J. A., Saloner, G., Salop, S. C., et al. (1990). Equilibrium vertical foreclosure.
American Economic Review, 80(1):127–142.
Orzen, H. (2008). Counterintuitive number effects in experimental oligopolies. Experimental Economics, 11(4):390–401.
O’Toole, G. (2011). Everything should be made as simple as possible, but not simpler. Available at http://quoteinvestigator.com/2011/05/13/einsteinsimple/. Accessed May 05, 2016.
Parente, P. M. and Silva, J. S. (2016). Quantile regression with clustered data. Journal of
Econometric Methods, 5(1):1–15.
Parker, G. and Van Alstyne, M. W. (2014). Innovation, openness, and platform control.
Working Paper. Available at http://ssrn.com/abstract=1079712.
332
References
Parker, G. G. and Van Alstyne, M. W. (2005). Two-sided network effects: A theory of
information product design. Management Science, 51(10):1494–1504.
Parker, P. M. and Röller, L.-H. (1997). Collusive conduct in duopolies: Multimarket
contact and cross-ownership in the mobile telephone industry. The RAND Journal of
Economics, 28(2):304–322.
Pashalidis, A. and Mitchell, C. J. (2003). A taxonomy of single sign-on systems. In
Safavi-Naini, R. and Seberry, J., editors, Information Security and Privacy, pages 249–
264. Springer, Berlin, Germany.
Peitz, M. and Valletti, T. (2015). Reassessing competition concerns in electronic communications markets. Telecommunications Policy, 39(10):896–912.
Pettit, J., Friedman, D., Kephart, C., and Oprea, R. (2014). Software for continuous game
experiments. Experimental Economics, 17(4):631–648.
Petulowa, M. and Saavedra, C. (2014). Margin squeeze in a regulatory environment:
An application to differentiated product markets. Working Paper. Available at http:
//ssrn.com/abstract=2236258.
Pew Research Center (2016).
Social networking fact sheet.
Available at
http://www.pewinternet.org/fact-sheets/social-networkingfact-sheet/. Accessed March 02, 2016.
Pittman, R. (2003). Vertical restructuring (or not) of the infrastructure sectors of transition economies. Journal of Industry, Competition and Trade, 3(1):5–26.
Plott, C. (1987). Dimensions of parallelism: Some policy applications of experimental
methods. In Roth, A., editor, Laboratory Experimentation in Economics: Six Points of
View. Cambridge University Press, New York, NY.
Potters, J. and Suetens, S. (2013). Oligopoly experiments in the current millennium.
Journal of Economic Surveys, 27(3):439–460.
333
References
Prinz, F., Schlange, T., and Asadullah, K. (2011). Believe it or not: how much can we
rely on published data on potential drug targets? Nature Reviews Drug Discovery,
10:712–712.
Radner, R. (1986).
Can bounded rationality resolve the prisoner’s dilemma?
In
Mas-Colell, A. and Hildenbrand, W., editors, Contributions to Mathematical Economics,
pages 387–399. North-Holland, Amsterdam, Netherlands.
Reisinger, M. (2012). Platform competition for advertisers and users in media markets.
International Journal of Industrial Organization, 30(2):243–252.
Reitzes, J. and Woroch, G. (2009). Input quality sabotage and the welfare consequences
of parity rules. In Telecommunications Policy Research Conference (TPRC). Arlington,
VA.
Renda, A. (2010). Competition–regulation interface in telecommunications: What’s left
of the essential facility doctrine. Telecommunications Policy, 34(1):23–35.
Renda, A. (2015).
Towards an empty connected continent package?
Available
at http://www.thedigitalpost.eu/2015/channel-telecoms/towardsan-empty-connected-continent-package. Accessed May 05, 2016.
Rey, P. and Tirole, J. (2007a). Financing and access in cooperatives. International Journal
of Industrial Organization, 25(5):1061–1088.
Rey, P. and Tirole, J. (2007b). A primer on foreclosure. In Armstrong, M. and Porter,
R. H., editors, Handbook of Industrial Organization, volume 3, pages 2145–2220. NorthHolland, Amsterdam, Netherlands.
Riordan, M. H. (2008). Competitive effects of vertical integration. In Buccirossi, P.,
editor, Handbook of Antitrust Economics. MIT Press, Cambridge, MA.
Robinson, J. (1962). Essays in the Theory of Economic Growth. Macmillan, London, UK.
334
References
Rochet, J.-C. and Tirole, J. (2006). Two-sided markets: a progress report. The RAND
Journal of Economics, 37(3):645–667.
Rosenau, P. V. (2000). Public-Private Policy Partnerships. MIT Press, Cambridge, MA.
Ross, P. (2015).
The shifting landscape of Facebook advertising.
Available at
http://www.socialbakers.com/blog/2401-the-shifting-landscapeof-facebook-advertising. Accessed March 02, 2016.
Roth, A. E. (1994). Lets keep the con out of experimental econ.: A methodological note.
Empirical Economics, 19(2):279–89.
Roth, A. E. (1995). Bargaining experiments. In Kagel, J. H. and Roth, A. E., editors,
The Handbook of Experimental Economics, pages 253–348. Princeton University Press,
Princeton, NJ.
Roth, A. E. (2002). The economist as engineer: Game theory, experimentation, and
computation as tools for design economics. Econometrica, 70(4):1341–1378.
Roth, A. E., Murnighan, J. K., and Schoumaker, F. (1988). The deadline effect in bargaining: Some experimental evidence. American Economic Review, 78(4):806–823.
Sacco, D. and Schmutzler, A. (2011). Is there a U-shaped relation between competition
and investment? International Journal of Industrial Organization, 29(1):65–73.
Salop, S. C. (1979). Monopolistic competition with outside goods. The Bell Journal of
Economics, 10(1):141–156.
Salop, S. C. and Scheffman, D. T. (1983). Raising rivals’ costs. American Economic Review,
73(2):267–271.
Salop, S. C. and Scheffman, D. T. (1987). Cost-raising strategies. The Journal of Industrial
Economics, 36(1):19–34.
335
References
Schellingerhout, R. (2011).
Standard-setting from a competition law perspec-
tive. Competition Policy Newsletter, 1:3–9. Available at http://ec.europa.eu/
competition/publications/cpn/2011_1_1_en.pdf.
Scherer, F. and Ross, D. (1990). Industrial Market Structure and Economic Performance.
Houghton Mifflin, Boston, MA, third edition.
Schmalensee, R. C. (1990). Empirical studies of rivalrous behavior. In Bonanno, G. and
Brandolini, D., editors, Industrial Structure in the New Industrial Economics. Clarendon
Press, Oxford, UK.
Schram, A. (2005). Artificiality: The tension between internal and external validity in
economic experiments. Journal of Economic Methodology, 12(2):225–237.
Schulz, K. F., Altman, D. G., and Moher, D. (2010). CONSORT 2010 Statement: Updated
guidelines for reporting parallel group randomised trials. PLoS Med, 7(3):1–7.
Schumpeter, J. A. (1942). Capitalism, Socialism, and Democracy. Harper and Brothers,
New York, NY.
Schwab, A., Abrahamson, E., Starbuck, W. H., and Fidler, F. (2011).
Perspective-
researchers should make thoughtful assessments instead of null-hypothesis significance tests. Organization Science, 22(4):1105–1120.
Selten, R. (1975). Reexamination of the perfectness concept for equilibrium points in
extensive games. International Journal of Game Theory, 4(1):25–55.
Shampanier, K., Mazar, N., and Ariely, D. (2007). Zero as a special price: The true value
of free products. Marketing Science, 26(6):742–757.
Shapiro, C. (1996). Mergers with differentiated products. Antitrust, 10(2):23–30.
Shleifer, A. (1998). State versus private ownership. Journal of Economic Perspectives,
12(4):133–150.
336
References
Shubik, M. and Levitan, R. (1980). Market Structure and Behavior. Harvard University
Press, Cambridge, MA.
Sibley, D. S. and Weisman, D. L. (1998). Raising rivals’ costs: The entry of an upstream
monopolist into downstream markets. Information Economics and Policy, 10(4):451–
470.
SimilarTech (2016). Technology: Facebook Connect. Available at https://www.
similartech.com/technologies/facebook-connect. Accessed March 02,
2016.
Simmons, J. P., Nelson, L. D., and Simonsohn, U. (2011). False-positive psychology:
Undisclosed flexibility in data collection and analysis allows presenting anything as
significant. Psychological Science, 22(11):1359–1366.
Simon, J. L. and Arndt, J. (1980). The shape of the advertising response function. Journal
of Advertising Research, 20(4):11–28.
Simon, L. K. and Stinchcombe, M. B. (1989). Extensive form games in continuous time:
Pure strategies. Econometrica, 57(5):1171–1214.
Singh, N. and Vives, X. (1984). Price and quantity competition in a differentiated
duopoly. The RAND Journal of Economics, 15(4):546–554.
Spengler, J. J. (1950). Vertical integration and antitrust policy. Journal of Political Economy,
58(4):347–352.
Speta, J. B. (2000). Handicapping the race for the last mile: A critque of open access
rules for broadband platforms. Yale Journal on Regulation, 17:39–91.
Steinbock, D. (2005). The Mobile Revolution: The Making of Mobile Services Worldwide.
Kogan Page, London, UK.
337
References
Stewart, J. B. (2013).
For airlines, it may be one merger too many.
The New
York Times. Available at http://www.nytimes.com/2013/08/17/business/
for-airlines-it-may-be-one-merger-too-many.html. Accessed March
29, 2016.
Suetens, S. and Potters, J. (2007). Bertrand colludes more than Cournot. Experimental
Economics, 10(1):71–77.
Sugden, R. (2000). Credible worlds: the status of theoretical models in economics. Journal of Economic Methodology, 7(1):1–31.
Sun, S.-T. and Beznosov, K. (2012). The devil is in the (implementation) details: An
empirical analysis of OAuth SSO systems. In Proceedings of the 2012 ACM conference
on Computer and communications security, pages 378–390. Raleigh, NC.
Sun, S.-T., Boshmaf, Y., Hawkey, K., and Beznosov, K. (2010). A billion keys, but few
locks: The crisis of web single sign-on. In Proceedings of the 2010 Workshop on New
Security Paradigms, pages 61–72. Concord, MA.
Sutton, R. I. and Staw, B. M. (1995). What theory is not. Administrative Science Quarterly,
40(3):371–384.
Taylor, C. R. (2004). Consumer privacy and the market for customer information. The
RAND Journal of Economics, 35(4):631–650.
Teppayayon, O. and Bohlin, E. (2010). Functional separation in Swedish broadband
market: Next step of improving competition. Telecommunications Policy, 34(7):375–
383.
Thirumalai, S. and Sinha, K. K. (2013). To personalize or not to personalize online
purchase interactions: Implications of self-selection by retailers. Information Systems
Research, 24(3):683–708.
338
References
Tropina, T., Whalley, J., and Curwen, P. (2010). Functional separation within the European Union: Debates and challenges. Telematics and Informatics, 27(3):231–241.
Troulos, C. and Maglaris, V. (2011). Factors determining municipal broadband strategies across Europe. Telecommunications Policy, 35(9):842–856.
Troulos, C., Merekoulias, V., and Maglaris, V. (2010). A business model for municipal
FTTH/B networks: the case of rural Greece. info, 12(3):73–89.
Tucker, C. E. (2012). The economics of advertising and privacy. International Journal of
Industrial Organization, 3(30):326–329.
Tucker, C. E. (2014). Social networks, personalized advertising, and privacy controls.
Journal of Marketing Research, 51(5):546–562.
Turow, J., King, J., Hoofnagle, C. J., Bleakley, A., and Hennessy, M. (2009). Americans
reject tailored advertising and three activities that enable it. Working Paper. Available
at http://ssrn.com/abstract=1478214.
Ünver, M. B. (2015). Is a fine-tuning approach sufficient for EU NGA policy? A global
review around the long-lasting debate. Telecommunications Policy, 39(11):957–979.
Urban, G. L., Liberali, G., MacDonald, E., Bordley, R., and Hauser, J. R. (2014). Morphing banner advertising. Marketing Science, 33(1):27–46.
Van Der Sype, Y. S. and Seigneur, J.-M. (2014). Case study: Legal requirements for the
use of social login features for online reputation updates. In Proceedings of the 29th
Annual ACM Symposium on Applied Computing, pages 1698–1705. Gyeongju, Republic
of Korea.
van Schewick, B. (2010). Internet Architecture and Innovation. MIT Press, Cambridge,
MA.
339
References
Vande Walle, S. and Wambach, J. (2014).
No magic number to dial - The com-
mission’s review of mobile telecom mergers. Competition Merger Brief, 2014(1):10–
16. Available at http://ec.europa.eu/competition/publications/cmb/
2014/CMB2014-01.pdf.
Vareda, J. (2011). Quality upgrades and bypass under mandatory access. Journal of
Regulatory Economics, 40(2):177–197.
Verdier, M. (2013). One sided access in two-sided markets. Working Paper. Available at
http://ssrn.com/abstract=2209379.
Vestager, M. (2015).
Competition in telecom markets.
Available at https:
//ec.europa.eu/commission/2014-2019/vestager/announcements/
competition-telecom-markets_en. Accessed May 05, 2016.
Vickers, J. (1995). Competition and regulation in vertically related markets. The Review
of Economic Studies, 62(1):1–17.
Viscusi, W. K., Harrington, J. E., and Vernon, J. M. (2005). Economics of Regulation and
Antitrust. MIT Press, Cambridge, MA, fourth edition.
Vogelsang, I. (2013). The endgame of telecommunications policy? A survey. Review of
Economics, 64(3):193–270.
Vogelstein, F. (2013). Dogfight: How Apple and Google Went to War and Started a Revolution.
Sarah Crichton Books/Farrar, Straus & Giroux, New York, NY.
Waichman, I., Requate, T., and Siang, C. K. (2014). Communication in Cournot competition: An experimental study. Journal of Economic Psychology, 42:1–16.
Waltman, L. and Kaymak, U. (2008). Q-learning agents in a Cournot oligopoly model.
Journal of Economic Dynamics and Control, 32(10):3275–3293.
340
References
Wang, R., Chen, S., and Wang, X. (2012). Signing me onto your accounts through Facebook and Google: a traffic-guided security study of commercially deployed singlesign-on web services. In IEEE Symposium on Security and Privacy, pages 365–379. San
Francisco, CA.
Weick, K. (2005). Definition of ‘theory’. In Nicholson, N., Audia, P., and Pillutla, M. M.,
editors, The Blackwell Encyclopedia of Management, Volume 11, Organizational Behavior.
Blackwell, Oxford, UK.
Weisman, D. L. (1995). Regulation and the vertically integrated firm: The case of RBOC
entry into interlata long distance. Journal of Regulatory Economics, 8(3):249–266.
Weisman, D. L. and Kang, J. (2001). Incentives for discrimination when upstream
monopolists participate in downstream markets. Journal of Regulatory Economics,
20(2):125–139.
West, J. (2003). How open is open enough?: Melding proprietary and open source
platform strategies. Research Policy, 32(7):1259–1285.
Whinston, M. D. (2003). On the transaction cost determinants of vertical integration.
Journal of Law, Economics, & Organization, 19(1):1–23.
Winther, R. G. (2015). The structure of scientific theories. In Zalta, E. N., editor, The
Stanford Encyclopedia of Philosophy. Center for the Study of Language and Information,
Stanford University, Stanford, CA, Fall 2015 edition.
Wong, E. (2012). Next-generation broadband access networks and technologies. Journal
of Lightwave Technology, 30(4):597–608.
Wong, S. (1973). The “F-twist” and the methodology of Paul Samuelson. American
Economic Review, 63(3):312–325.
Wright, J. et al. (2004). One-sided logic in two-sided markets. Review of Network Economics, 3(1):1–21.
341
References
Yescombe, E. R. (2011).
Public-private Partnerships: Principles of Policy and Finance.
Butterworth-Heinemann, Burlington, MA.
Yoo, C. S. (2002). Vertical integration and media regulation in the new economy. Yale
Journal on Regulation, 19:171–300.
Yoo, C. Y. (2009). Effects beyond click-through: Incidental exposure to web advertising.
Journal of Marketing Communications, 15(4):227–246.
342
List of Abbreviations
API
application programming interface
CLEC
competitive local exchange carriers
CP
content provider
CPA
cost-per-action
CPC
cost-per-click
CPM
cost-per-mille
EU
European Union
FBC
facility-based competition
ICT
information and communications technology
IO
Industrial Organization
IP
Internet Protocol
IS
Information Systems
JPM
joint profit maximization
LRIC
long-run incremental costs
MSR
margin squeeze regulation
343
List of Abbreviations
NGA
next-generation access
NGAN
next-generation access network
NR
no regulation
NRA
national regulatory authority
OA
Open Access
OSI
Open Systems Interconnection
pp
percentage points
PPP
public-private-partnership
QoS
quality of service
RRC
raising rivals’ costs
SBC
service-based competition
SMPT
Single Monopoly Profit Theorem
SOE
state-owned enterprise
WC
Wholesale Competition
WCPC
Wholesale Competition with Price Commitment
WM
Wholesale Monopoly
344
List of Figures
2.1
OA classification framework: vertical structure (impact on downstream
market), ownership (goal of organization), and access level (quality differentiation). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
20
2.2
Classification of OA application scenarios. . . . . . . . . . . . . . . . . . .
22
2.3
Recommended policy guideline for OA regulation of NGANs. . . . . . .
46
3.1
Idealized microeconomically founded IS research process cycle. . . . . .
60
3.2
Normalized degree of tacit collusion based on prices relative to the Nash
equilibrium (based on Engel, 2007, p.494). . . . . . . . . . . . . . . . . . .
93
4.1
Market structure (based on Bourreau et al., 2011, p. 683). . . . . . . . . . 106
4.2
Equilibrium prices under NR and MSR. . . . . . . . . . . . . . . . . . . . 109
4.3
MSR effect on market variables in the scenario with a competitive retail
fringe. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
4.4
Equilibrium prices under NR and MSR. . . . . . . . . . . . . . . . . . . . 112
4.5
MSR effect on market variables in the scenario with simultaneous retail
pricing. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
5.1
Conceptual model of wholesale competition with two integrated firms
and a non-integrated retailer (based on Bourreau et al., 2011, p. 683). . . 129
5.2
Median wholesale (dashed) and retail (solid) market prices for each of the
six treatments. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
5.3
Median wholesale (dashed) and retail (solid) market prices of students and
experts. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
345
List of Figures
6.1
Average degrees of tacit collusion j Nash over periods across treatments.
6.2
Number effects across symmetric treatments. . . . . . . . . . . . . . . . . 182
7.1
The competitive setting of the model: Each user chooses exactly one of
171
the two special-interest CPs (single-homing) and splits attention between
the special-interest CP and the general-interest CP (multi-homing). All
CPs are ad-financed and receive advertisement revenues according to
their relative ability to reach ad-relevant users. . . . . . . . . . . . . . . . 195
7.2
Illustration of possible market outcomes: The social login is offered in
outcomes I, I I and IV and not offered otherwise. The social login is
adopted in outcomes I and I I and not adopted in outcome IV. In outcome I special-interest CPs are better off and in outcome I I they are
worse off by adopting the social login. Note: The figure is derived for
abs = 0.5, abG(s) = 0.5, t = 0.5, q = 0.1, d = 0.5. Not all market outcomes
may exist for all feasible parameter constellations. . . . . . . . . . . . . . 207
A.1 Comparison of profits in the case of (non-)foreclosure under NR. . . . . 238
A.2 Comparison of Firm A’s profit in the case of (non-)foreclosure under NR. 239
A.3 MSR effect on consumer surplus, producer surplus, and total surplus. . 242
B.1 Period averages of wholesale (dashed) and retail (solid) market prices.
Graphs are plotted based on medians of 50 ticks. . . . . . . . . . . . . . . 264
B.2 Average degrees of tacit collusion jWalras over periods across treatments. 272
C.1 Display of the experimental software. . . . . . . . . . . . . . . . . . . . . 275
C.2 Structure of the experiment. . . . . . . . . . . . . . . . . . . . . . . . . . . 279
C.3 Display of the experimental software. . . . . . . . . . . . . . . . . . . . . 281
C.4 Display of the experimental software. . . . . . . . . . . . . . . . . . . . . 285
346
List of Tables
2.1
Summary - Vertical Integration & Separation. . . . . . . . . . . . . . . . .
33
2.2
Summary - Co-investment. . . . . . . . . . . . . . . . . . . . . . . . . . .
37
2.3
Summary - Public-sector participation.
. . . . . . . . . . . . . . . . . . .
44
3.1
Economic laboratory experiments in non-discrete time. . . . . . . . . . .
84
3.2
Treatment variables and their values. . . . . . . . . . . . . . . . . . . . . .
87
3.3
Scaled theoretical benchmarks of oligopoly competition for each treatment as displayed in the experiment. . . . . . . . . . . . . . . . . . . . . .
89
3.4
Average degrees of tacit collusion across treatments. . . . . . . . . . . . .
95
3.5
Multilevel mixed-effects linear regressions of the degree of tacit collusion
on treatments. Baseline: Discrete Bertrand Duopoly (DB2). . . . . . . . .
98
5.1
Predicted wholesale and retail price differences between market scenarios.132
5.2
Full-factorial experimental design with six treatments. . . . . . . . . . . . 135
5.3
Treatment median of median market cohort prices, profits and consumer
surplus. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
5.4
Quantile regressions of wholesale market price am , retail market price
ym , integrated firms’ average profit p AB , retailer’s profit p D and conf Baseline: Wholesale Monopoly & No Regulation
sumer surplus CS.
(WM-NR). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140
5.5
Quantile regressions of market outcomes in case of wholesale competition without price commitment. Baseline: Wholesale Competition & No
Regulation (WC-NR). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
347
List of Tables
5.6
Quantile regressions of market outcomes in case of wholesale competition with price commitment. Baseline: Wholesale Competition with
Price Commitment & No Regulation (WCPC-NR). . . . . . . . . . . . . . 144
5.7
Quantile regressions of market outcomes on subject type in case of
wholesale competition with price commitment. Baseline: Wholesale
Competition with Price Commitment & No Regulation (WCPC-NR) with
Students. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146
5.8
Profits and discount factors under wholesale monopoly. . . . . . . . . . . 148
5.9
Profits and discount factors under wholesale competition. . . . . . . . . 148
6.1
Economic laboratory experiments that vary the number of competing
firms. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158
6.2
Degrees of tacit collusion in economic laboratory experiments that vary
the number of competing firms. . . . . . . . . . . . . . . . . . . . . . . . . 161
6.3
Intra-study one-tailed Mann-Whitney U tests and associated p values. . 162
6.4
Multilevel mixed-effects linear regressions of tacit collusion on number
of competitors and competition model on the basis of most comparable
treatments. Baseline: Triopoly & Bertrand competition. . . . . . . . . . . 164
6.5
Inter-study average degrees of tacit collusion and one-tailed matchedsamples Wilcoxon signed-rank tests on the basis of most comparable
treatments. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165
6.6
Nash predictions p Nash /q Nash as measured by the Walrasian-based degrees of tacit collusion jWalras . . . . . . . . . . . . . . . . . . . . . . . . . . 169
6.7
Average degrees of tacit collusion across treatments. . . . . . . . . . . . . 171
6.8
Multilevel mixed-effects linear regressions of tacit collusion on number
of competitors and competition model under competition between symmetric firms. Baseline: Symmetric Bertrand Triopoly (B3).
6.9
. . . . . . . . 173
Average degrees of tacit collusion across asymmetry treatments. . . . . . 177
6.10 Multilevel mixed-effects linear regressions of tacit collusion on number
of competitors under Bertrand competition between asymmetric firms.
Baseline: Asymmetric Bertrand Triopoly (B3A).
348
. . . . . . . . . . . . . . 179
List of Tables
7.1
Normal form game representing special-interest CPs’ social login adoption decision.
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203
A.1 Scaled theoretical benchmarks of oligopoly competition for each treatment with symmetric firm as displayed in the experiment. . . . . . .
234
A.2 Scaled theoretical benchmarks of oligopoly competition for the asymmetric Bertrand treatments as displayed in the experiment. . . . . . .
235
A.3 Nash predictions p Nash as measured by Walrasian-based degree of tacit
collusion jWalras under asymmetric Bertrand competition controlling for
scaling in each treatment. . . . . . . . . . . . . . . . . . . . . . . . . .
235
A.4 Theoretical predictions for Timing Model (1). . . . . . . . . . . . . . .
245
A.5 Theoretical predictions for Timing Model (2). . . . . . . . . . . . . . .
247
A.6 Theoretical predictions for Timing Model (3). . . . . . . . . . . . . . .
248
A.7 Theoretical predictions for Timing Model (4). . . . . . . . . . . . . . .
249
B.1 Average degrees of tacit collusion over the entire time horizon across
treatments. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
259
B.2 Quantile regression of integrated firms’ profit pi in the case of a wholesale monopoly. Baseline: Firm B in Wholesale Monopoly under No Regulation (WM-NR). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
260
B.3 Quantile regression of wholesale market prices am , retail market prices
ym , integrated firms’ average profits p AB , retailer’s profits p D and conf in the case of a wholesale monopoly. Baseline: Wholesumer surplus CS
sale Monopoly & No Regulation (WM-NR). . . . . . . . . . . . . . . .
261
B.4 Pairwise treatment effects on the wholesale market price am . . . . . .
261
B.5 Pairwise treatment effects on the retail market price ym . . . . . . . . .
262
B.6 Pairwise treatment effects on the average profit of integrated firms p AB .
262
B.7 Pairwise treatment effects on the retailer’s profit p D . . . . . . . . . . .
262
f . . . . . . . . . .
B.8 Pairwise treatment effects on consumer surplus CS.
263
B.9 Average market prices, profits and consumer surplus per treatment.
263
349
List of Tables
B.10 Linear regression of wholesale market prices am , retail market prices ym ,
integrated firms’ average profits p AB , retailer’s profits p D and consumer
f Baseline: Wholesale Monopoly under No Regulation (WMsurplus CS.
NR). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
265
B.11 Meta-regression of tacit collusion on number of competitors and competition model on the basis of most comparable treatments. Baseline:
Triopoly & Bertrand competition.
. . . . . . . . . . . . . . . . . . . .
267
B.12 Multilevel mixed-effects linear regressions of tacit collusion on
(a)symmetry of firms in triopolies. Baseline: Symmetric Bertrand Triopoly (B3). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
268
B.13 Multilevel mixed-effects linear regressions of tacit collusion on
(a)symmetry of firms in quadropolies. Baseline: Symmetric Bertrand
Quadropoly (B4). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
269
B.14 Multilevel mixed-effects linear regressions of tacit collusion on number
of competitors and competition model on the basis of all treatments.
Baseline: Triopoly & Bertrand competition. . . . . . . . . . . . . . . .
271
B.15 Inter-study average degrees of tacit collusion and one-tailed matchedsamples Wilcoxon signed-rank tests on the basis of all treatments. . .
350
271